Instructions to use openbmb/MiniCPM-V-4-gguf with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use openbmb/MiniCPM-V-4-gguf with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="openbmb/MiniCPM-V-4-gguf", trust_remote_code=True)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("openbmb/MiniCPM-V-4-gguf", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use openbmb/MiniCPM-V-4-gguf with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf openbmb/MiniCPM-V-4-gguf:Q4_K_M

Use Docker

docker model run hf.co/openbmb/MiniCPM-V-4-gguf:Q4_K_M

LM Studio
Jan

vLLM

How to use openbmb/MiniCPM-V-4-gguf with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "openbmb/MiniCPM-V-4-gguf"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM-V-4-gguf",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/openbmb/MiniCPM-V-4-gguf:Q4_K_M

SGLang

How to use openbmb/MiniCPM-V-4-gguf with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "openbmb/MiniCPM-V-4-gguf" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM-V-4-gguf",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "openbmb/MiniCPM-V-4-gguf" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "openbmb/MiniCPM-V-4-gguf",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Ollama
How to use openbmb/MiniCPM-V-4-gguf with Ollama:
```
ollama run hf.co/openbmb/MiniCPM-V-4-gguf:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use openbmb/MiniCPM-V-4-gguf with Docker Model Runner:
```
docker model run hf.co/openbmb/MiniCPM-V-4-gguf:Q4_K_M
```

Lemonade

How to use openbmb/MiniCPM-V-4-gguf with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull openbmb/MiniCPM-V-4-gguf:Q4_K_M

Run and chat with the model

lemonade run user.MiniCPM-V-4-gguf-Q4_K_M

List all available models

lemonade list

Atomic Chat

MiniCPM-V-4-gguf

File size: 330,682 Bytes

f174180

program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.6.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
{
    func main<ios15>(tensor<fp32, [1, 1024, 1152]> input) {
            tensor<int32, []> var_11 = const()[name = tensor<string, []>("op_11"), val = tensor<int32, []>(-1)];
            tensor<int32, [1]> hidden_states_1_axes_0 = const()[name = tensor<string, []>("hidden_states_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<string, []> input_to_fp16_dtype_0 = const()[name = tensor<string, []>("input_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1152]> encoder_layers_0_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
            tensor<fp16, [1152]> encoder_layers_0_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))];
            tensor<fp16, []> var_5_to_fp16 = const()[name = tensor<string, []>("op_5_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 1024, 1152]> input_to_fp16 = cast(dtype = input_to_fp16_dtype_0, x = input)[name = tensor<string, []>("cast_136")];
            tensor<fp16, [1, 1024, 1152]> hidden_states_1_cast_fp16 = layer_norm(axes = hidden_states_1_axes_0, beta = encoder_layers_0_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_layer_norm1_weight_to_fp16, x = input_to_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4800)))];
            tensor<fp16, [1152]> encoder_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2659072)))];
            tensor<fp16, [1, 1024, 1152]> linear_0_cast_fp16 = linear(bias = encoder_layers_0_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_q_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2661440)))];
            tensor<fp16, [1152]> encoder_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5315712)))];
            tensor<fp16, [1, 1024, 1152]> linear_1_cast_fp16 = linear(bias = encoder_layers_0_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_k_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5318080)))];
            tensor<fp16, [1152]> encoder_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7972352)))];
            tensor<fp16, [1, 1024, 1152]> linear_2_cast_fp16 = linear(bias = encoder_layers_0_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_v_proj_weight_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> var_96 = const()[name = tensor<string, []>("op_96"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_97_cast_fp16 = reshape(shape = var_96, x = linear_0_cast_fp16)[name = tensor<string, []>("op_97_cast_fp16")];
            tensor<int32, [4]> var_99 = const()[name = tensor<string, []>("op_99"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_100_cast_fp16 = reshape(shape = var_99, x = linear_1_cast_fp16)[name = tensor<string, []>("op_100_cast_fp16")];
            tensor<int32, [4]> var_102 = const()[name = tensor<string, []>("op_102"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_103_cast_fp16 = reshape(shape = var_102, x = linear_2_cast_fp16)[name = tensor<string, []>("op_103_cast_fp16")];
            tensor<int32, [4]> value_states_3_perm_0 = const()[name = tensor<string, []>("value_states_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_106_transpose_x_0 = const()[name = tensor<string, []>("op_106_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_106_transpose_y_0 = const()[name = tensor<string, []>("op_106_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_82 = transpose(perm = transpose_82_perm_0, x = var_100_cast_fp16)[name = tensor<string, []>("transpose_240")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_81 = transpose(perm = transpose_81_perm_0, x = var_97_cast_fp16)[name = tensor<string, []>("transpose_241")];
            tensor<fp16, [1, 16, 1024, 1024]> var_106_cast_fp16 = matmul(transpose_x = var_106_transpose_x_0, transpose_y = var_106_transpose_y_0, x = transpose_81, y = transpose_82)[name = tensor<string, []>("op_106_cast_fp16")];
            tensor<fp16, []> var_107_to_fp16 = const()[name = tensor<string, []>("op_107_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_1_cast_fp16 = mul(x = var_106_cast_fp16, y = var_107_to_fp16)[name = tensor<string, []>("attn_weights_1_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_109_cast_fp16 = softmax(axis = var_11, x = attn_weights_1_cast_fp16)[name = tensor<string, []>("op_109_cast_fp16")];
            tensor<bool, []> attn_output_1_transpose_x_0 = const()[name = tensor<string, []>("attn_output_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_1_transpose_y_0 = const()[name = tensor<string, []>("attn_output_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_3_cast_fp16 = transpose(perm = value_states_3_perm_0, x = var_103_cast_fp16)[name = tensor<string, []>("transpose_242")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_1_cast_fp16 = matmul(transpose_x = attn_output_1_transpose_x_0, transpose_y = attn_output_1_transpose_y_0, x = var_109_cast_fp16, y = value_states_3_cast_fp16)[name = tensor<string, []>("attn_output_1_cast_fp16")];
            tensor<int32, [4]> var_113_perm_0 = const()[name = tensor<string, []>("op_113_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_115 = const()[name = tensor<string, []>("op_115"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_113_cast_fp16 = transpose(perm = var_113_perm_0, x = attn_output_1_cast_fp16)[name = tensor<string, []>("transpose_239")];
            tensor<fp16, [1, 1024, 1152]> input_3_cast_fp16 = reshape(shape = var_115, x = var_113_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7974720)))];
            tensor<fp16, [1152]> encoder_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10628992)))];
            tensor<fp16, [1, 1024, 1152]> linear_3_cast_fp16 = linear(bias = encoder_layers_0_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_0_self_attn_out_proj_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_5_cast_fp16 = add(x = input_to_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<int32, [1]> input_7_axes_0 = const()[name = tensor<string, []>("input_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_0_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10631360)))];
            tensor<fp16, [1152]> encoder_layers_0_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10633728)))];
            tensor<fp16, [1, 1024, 1152]> input_7_cast_fp16 = layer_norm(axes = input_7_axes_0, beta = encoder_layers_0_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_layer_norm2_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_0_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10636096)))];
            tensor<fp16, [4304]> encoder_layers_0_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20552576)))];
            tensor<fp16, [1, 1024, 4304]> linear_4_cast_fp16 = linear(bias = encoder_layers_0_mlp_fc1_bias_to_fp16, weight = encoder_layers_0_mlp_fc1_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<string, []> input_11_mode_0 = const()[name = tensor<string, []>("input_11_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_11_cast_fp16 = gelu(mode = input_11_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_0_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20561280)))];
            tensor<fp16, [1152]> encoder_layers_0_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30477760)))];
            tensor<fp16, [1, 1024, 1152]> linear_5_cast_fp16 = linear(bias = encoder_layers_0_mlp_fc2_bias_to_fp16, weight = encoder_layers_0_mlp_fc2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_13_cast_fp16 = add(x = input_5_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<int32, [1]> hidden_states_7_axes_0 = const()[name = tensor<string, []>("hidden_states_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_1_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30480128)))];
            tensor<fp16, [1152]> encoder_layers_1_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30482496)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_7_cast_fp16 = layer_norm(axes = hidden_states_7_axes_0, beta = encoder_layers_1_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_layer_norm1_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30484864)))];
            tensor<fp16, [1152]> encoder_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33139136)))];
            tensor<fp16, [1, 1024, 1152]> linear_6_cast_fp16 = linear(bias = encoder_layers_1_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_q_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33141504)))];
            tensor<fp16, [1152]> encoder_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35795776)))];
            tensor<fp16, [1, 1024, 1152]> linear_7_cast_fp16 = linear(bias = encoder_layers_1_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_k_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35798144)))];
            tensor<fp16, [1152]> encoder_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38452416)))];
            tensor<fp16, [1, 1024, 1152]> linear_8_cast_fp16 = linear(bias = encoder_layers_1_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_v_proj_weight_to_fp16, x = hidden_states_7_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<int32, [4]> var_158 = const()[name = tensor<string, []>("op_158"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_159_cast_fp16 = reshape(shape = var_158, x = linear_6_cast_fp16)[name = tensor<string, []>("op_159_cast_fp16")];
            tensor<int32, [4]> var_161 = const()[name = tensor<string, []>("op_161"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_162_cast_fp16 = reshape(shape = var_161, x = linear_7_cast_fp16)[name = tensor<string, []>("op_162_cast_fp16")];
            tensor<int32, [4]> var_164 = const()[name = tensor<string, []>("op_164"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_165_cast_fp16 = reshape(shape = var_164, x = linear_8_cast_fp16)[name = tensor<string, []>("op_165_cast_fp16")];
            tensor<int32, [4]> value_states_7_perm_0 = const()[name = tensor<string, []>("value_states_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_168_transpose_x_0 = const()[name = tensor<string, []>("op_168_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_168_transpose_y_0 = const()[name = tensor<string, []>("op_168_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_84 = transpose(perm = transpose_84_perm_0, x = var_162_cast_fp16)[name = tensor<string, []>("transpose_236")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_83 = transpose(perm = transpose_83_perm_0, x = var_159_cast_fp16)[name = tensor<string, []>("transpose_237")];
            tensor<fp16, [1, 16, 1024, 1024]> var_168_cast_fp16 = matmul(transpose_x = var_168_transpose_x_0, transpose_y = var_168_transpose_y_0, x = transpose_83, y = transpose_84)[name = tensor<string, []>("op_168_cast_fp16")];
            tensor<fp16, []> var_169_to_fp16 = const()[name = tensor<string, []>("op_169_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_5_cast_fp16 = mul(x = var_168_cast_fp16, y = var_169_to_fp16)[name = tensor<string, []>("attn_weights_5_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_171_cast_fp16 = softmax(axis = var_11, x = attn_weights_5_cast_fp16)[name = tensor<string, []>("op_171_cast_fp16")];
            tensor<bool, []> attn_output_5_transpose_x_0 = const()[name = tensor<string, []>("attn_output_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_5_transpose_y_0 = const()[name = tensor<string, []>("attn_output_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_7_cast_fp16 = transpose(perm = value_states_7_perm_0, x = var_165_cast_fp16)[name = tensor<string, []>("transpose_238")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_5_cast_fp16 = matmul(transpose_x = attn_output_5_transpose_x_0, transpose_y = attn_output_5_transpose_y_0, x = var_171_cast_fp16, y = value_states_7_cast_fp16)[name = tensor<string, []>("attn_output_5_cast_fp16")];
            tensor<int32, [4]> var_175_perm_0 = const()[name = tensor<string, []>("op_175_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_177 = const()[name = tensor<string, []>("op_177"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_175_cast_fp16 = transpose(perm = var_175_perm_0, x = attn_output_5_cast_fp16)[name = tensor<string, []>("transpose_235")];
            tensor<fp16, [1, 1024, 1152]> input_17_cast_fp16 = reshape(shape = var_177, x = var_175_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38454784)))];
            tensor<fp16, [1152]> encoder_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41109056)))];
            tensor<fp16, [1, 1024, 1152]> linear_9_cast_fp16 = linear(bias = encoder_layers_1_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_1_self_attn_out_proj_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_19_cast_fp16 = add(x = input_13_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")];
            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_1_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41111424)))];
            tensor<fp16, [1152]> encoder_layers_1_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41113792)))];
            tensor<fp16, [1, 1024, 1152]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = encoder_layers_1_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_layer_norm2_weight_to_fp16, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_1_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41116160)))];
            tensor<fp16, [4304]> encoder_layers_1_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51032640)))];
            tensor<fp16, [1, 1024, 4304]> linear_10_cast_fp16 = linear(bias = encoder_layers_1_mlp_fc1_bias_to_fp16, weight = encoder_layers_1_mlp_fc1_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<string, []> input_25_mode_0 = const()[name = tensor<string, []>("input_25_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_25_cast_fp16 = gelu(mode = input_25_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_1_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51041344)))];
            tensor<fp16, [1152]> encoder_layers_1_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60957824)))];
            tensor<fp16, [1, 1024, 1152]> linear_11_cast_fp16 = linear(bias = encoder_layers_1_mlp_fc2_bias_to_fp16, weight = encoder_layers_1_mlp_fc2_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_27_cast_fp16 = add(x = input_19_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")];
            tensor<int32, [1]> hidden_states_13_axes_0 = const()[name = tensor<string, []>("hidden_states_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_2_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60960192)))];
            tensor<fp16, [1152]> encoder_layers_2_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60962560)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_13_cast_fp16 = layer_norm(axes = hidden_states_13_axes_0, beta = encoder_layers_2_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_layer_norm1_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60964928)))];
            tensor<fp16, [1152]> encoder_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63619200)))];
            tensor<fp16, [1, 1024, 1152]> linear_12_cast_fp16 = linear(bias = encoder_layers_2_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_q_proj_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63621568)))];
            tensor<fp16, [1152]> encoder_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66275840)))];
            tensor<fp16, [1, 1024, 1152]> linear_13_cast_fp16 = linear(bias = encoder_layers_2_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_k_proj_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66278208)))];
            tensor<fp16, [1152]> encoder_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68932480)))];
            tensor<fp16, [1, 1024, 1152]> linear_14_cast_fp16 = linear(bias = encoder_layers_2_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_v_proj_weight_to_fp16, x = hidden_states_13_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_220 = const()[name = tensor<string, []>("op_220"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_221_cast_fp16 = reshape(shape = var_220, x = linear_12_cast_fp16)[name = tensor<string, []>("op_221_cast_fp16")];
            tensor<int32, [4]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_224_cast_fp16 = reshape(shape = var_223, x = linear_13_cast_fp16)[name = tensor<string, []>("op_224_cast_fp16")];
            tensor<int32, [4]> var_226 = const()[name = tensor<string, []>("op_226"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_227_cast_fp16 = reshape(shape = var_226, x = linear_14_cast_fp16)[name = tensor<string, []>("op_227_cast_fp16")];
            tensor<int32, [4]> value_states_11_perm_0 = const()[name = tensor<string, []>("value_states_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_230_transpose_x_0 = const()[name = tensor<string, []>("op_230_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_230_transpose_y_0 = const()[name = tensor<string, []>("op_230_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_86 = transpose(perm = transpose_86_perm_0, x = var_224_cast_fp16)[name = tensor<string, []>("transpose_232")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_85 = transpose(perm = transpose_85_perm_0, x = var_221_cast_fp16)[name = tensor<string, []>("transpose_233")];
            tensor<fp16, [1, 16, 1024, 1024]> var_230_cast_fp16 = matmul(transpose_x = var_230_transpose_x_0, transpose_y = var_230_transpose_y_0, x = transpose_85, y = transpose_86)[name = tensor<string, []>("op_230_cast_fp16")];
            tensor<fp16, []> var_231_to_fp16 = const()[name = tensor<string, []>("op_231_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_9_cast_fp16 = mul(x = var_230_cast_fp16, y = var_231_to_fp16)[name = tensor<string, []>("attn_weights_9_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_233_cast_fp16 = softmax(axis = var_11, x = attn_weights_9_cast_fp16)[name = tensor<string, []>("op_233_cast_fp16")];
            tensor<bool, []> attn_output_9_transpose_x_0 = const()[name = tensor<string, []>("attn_output_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_9_transpose_y_0 = const()[name = tensor<string, []>("attn_output_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_11_cast_fp16 = transpose(perm = value_states_11_perm_0, x = var_227_cast_fp16)[name = tensor<string, []>("transpose_234")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_9_cast_fp16 = matmul(transpose_x = attn_output_9_transpose_x_0, transpose_y = attn_output_9_transpose_y_0, x = var_233_cast_fp16, y = value_states_11_cast_fp16)[name = tensor<string, []>("attn_output_9_cast_fp16")];
            tensor<int32, [4]> var_237_perm_0 = const()[name = tensor<string, []>("op_237_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_239 = const()[name = tensor<string, []>("op_239"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_237_cast_fp16 = transpose(perm = var_237_perm_0, x = attn_output_9_cast_fp16)[name = tensor<string, []>("transpose_231")];
            tensor<fp16, [1, 1024, 1152]> input_31_cast_fp16 = reshape(shape = var_239, x = var_237_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68934848)))];
            tensor<fp16, [1152]> encoder_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71589120)))];
            tensor<fp16, [1, 1024, 1152]> linear_15_cast_fp16 = linear(bias = encoder_layers_2_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_2_self_attn_out_proj_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_33_cast_fp16 = add(x = input_27_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")];
            tensor<int32, [1]> input_35_axes_0 = const()[name = tensor<string, []>("input_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_2_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71591488)))];
            tensor<fp16, [1152]> encoder_layers_2_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71593856)))];
            tensor<fp16, [1, 1024, 1152]> input_35_cast_fp16 = layer_norm(axes = input_35_axes_0, beta = encoder_layers_2_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_layer_norm2_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_2_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71596224)))];
            tensor<fp16, [4304]> encoder_layers_2_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81512704)))];
            tensor<fp16, [1, 1024, 4304]> linear_16_cast_fp16 = linear(bias = encoder_layers_2_mlp_fc1_bias_to_fp16, weight = encoder_layers_2_mlp_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<string, []> input_39_mode_0 = const()[name = tensor<string, []>("input_39_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_2_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81521408)))];
            tensor<fp16, [1152]> encoder_layers_2_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91437888)))];
            tensor<fp16, [1, 1024, 1152]> linear_17_cast_fp16 = linear(bias = encoder_layers_2_mlp_fc2_bias_to_fp16, weight = encoder_layers_2_mlp_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_41_cast_fp16 = add(x = input_33_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")];
            tensor<int32, [1]> hidden_states_19_axes_0 = const()[name = tensor<string, []>("hidden_states_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_3_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91440256)))];
            tensor<fp16, [1152]> encoder_layers_3_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91442624)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_19_cast_fp16 = layer_norm(axes = hidden_states_19_axes_0, beta = encoder_layers_3_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_layer_norm1_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91444992)))];
            tensor<fp16, [1152]> encoder_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94099264)))];
            tensor<fp16, [1, 1024, 1152]> linear_18_cast_fp16 = linear(bias = encoder_layers_3_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_q_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94101632)))];
            tensor<fp16, [1152]> encoder_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96755904)))];
            tensor<fp16, [1, 1024, 1152]> linear_19_cast_fp16 = linear(bias = encoder_layers_3_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_k_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96758272)))];
            tensor<fp16, [1152]> encoder_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99412544)))];
            tensor<fp16, [1, 1024, 1152]> linear_20_cast_fp16 = linear(bias = encoder_layers_3_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_v_proj_weight_to_fp16, x = hidden_states_19_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> var_282 = const()[name = tensor<string, []>("op_282"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_283_cast_fp16 = reshape(shape = var_282, x = linear_18_cast_fp16)[name = tensor<string, []>("op_283_cast_fp16")];
            tensor<int32, [4]> var_285 = const()[name = tensor<string, []>("op_285"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_286_cast_fp16 = reshape(shape = var_285, x = linear_19_cast_fp16)[name = tensor<string, []>("op_286_cast_fp16")];
            tensor<int32, [4]> var_288 = const()[name = tensor<string, []>("op_288"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_289_cast_fp16 = reshape(shape = var_288, x = linear_20_cast_fp16)[name = tensor<string, []>("op_289_cast_fp16")];
            tensor<int32, [4]> value_states_15_perm_0 = const()[name = tensor<string, []>("value_states_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_292_transpose_x_0 = const()[name = tensor<string, []>("op_292_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_292_transpose_y_0 = const()[name = tensor<string, []>("op_292_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_88 = transpose(perm = transpose_88_perm_0, x = var_286_cast_fp16)[name = tensor<string, []>("transpose_228")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_87 = transpose(perm = transpose_87_perm_0, x = var_283_cast_fp16)[name = tensor<string, []>("transpose_229")];
            tensor<fp16, [1, 16, 1024, 1024]> var_292_cast_fp16 = matmul(transpose_x = var_292_transpose_x_0, transpose_y = var_292_transpose_y_0, x = transpose_87, y = transpose_88)[name = tensor<string, []>("op_292_cast_fp16")];
            tensor<fp16, []> var_293_to_fp16 = const()[name = tensor<string, []>("op_293_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_13_cast_fp16 = mul(x = var_292_cast_fp16, y = var_293_to_fp16)[name = tensor<string, []>("attn_weights_13_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_295_cast_fp16 = softmax(axis = var_11, x = attn_weights_13_cast_fp16)[name = tensor<string, []>("op_295_cast_fp16")];
            tensor<bool, []> attn_output_13_transpose_x_0 = const()[name = tensor<string, []>("attn_output_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_13_transpose_y_0 = const()[name = tensor<string, []>("attn_output_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_15_cast_fp16 = transpose(perm = value_states_15_perm_0, x = var_289_cast_fp16)[name = tensor<string, []>("transpose_230")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_13_cast_fp16 = matmul(transpose_x = attn_output_13_transpose_x_0, transpose_y = attn_output_13_transpose_y_0, x = var_295_cast_fp16, y = value_states_15_cast_fp16)[name = tensor<string, []>("attn_output_13_cast_fp16")];
            tensor<int32, [4]> var_299_perm_0 = const()[name = tensor<string, []>("op_299_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_301 = const()[name = tensor<string, []>("op_301"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_299_cast_fp16 = transpose(perm = var_299_perm_0, x = attn_output_13_cast_fp16)[name = tensor<string, []>("transpose_227")];
            tensor<fp16, [1, 1024, 1152]> input_45_cast_fp16 = reshape(shape = var_301, x = var_299_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99414912)))];
            tensor<fp16, [1152]> encoder_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102069184)))];
            tensor<fp16, [1, 1024, 1152]> linear_21_cast_fp16 = linear(bias = encoder_layers_3_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_3_self_attn_out_proj_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_47_cast_fp16 = add(x = input_41_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<int32, [1]> input_49_axes_0 = const()[name = tensor<string, []>("input_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_3_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102071552)))];
            tensor<fp16, [1152]> encoder_layers_3_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102073920)))];
            tensor<fp16, [1, 1024, 1152]> input_49_cast_fp16 = layer_norm(axes = input_49_axes_0, beta = encoder_layers_3_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_layer_norm2_weight_to_fp16, x = input_47_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_3_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102076288)))];
            tensor<fp16, [4304]> encoder_layers_3_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(111992768)))];
            tensor<fp16, [1, 1024, 4304]> linear_22_cast_fp16 = linear(bias = encoder_layers_3_mlp_fc1_bias_to_fp16, weight = encoder_layers_3_mlp_fc1_weight_to_fp16, x = input_49_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<string, []> input_53_mode_0 = const()[name = tensor<string, []>("input_53_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_53_cast_fp16 = gelu(mode = input_53_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_3_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112001472)))];
            tensor<fp16, [1152]> encoder_layers_3_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121917952)))];
            tensor<fp16, [1, 1024, 1152]> linear_23_cast_fp16 = linear(bias = encoder_layers_3_mlp_fc2_bias_to_fp16, weight = encoder_layers_3_mlp_fc2_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_55_cast_fp16 = add(x = input_47_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")];
            tensor<int32, [1]> hidden_states_25_axes_0 = const()[name = tensor<string, []>("hidden_states_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_4_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121920320)))];
            tensor<fp16, [1152]> encoder_layers_4_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121922688)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_25_cast_fp16 = layer_norm(axes = hidden_states_25_axes_0, beta = encoder_layers_4_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_layer_norm1_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121925056)))];
            tensor<fp16, [1152]> encoder_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124579328)))];
            tensor<fp16, [1, 1024, 1152]> linear_24_cast_fp16 = linear(bias = encoder_layers_4_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_q_proj_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124581696)))];
            tensor<fp16, [1152]> encoder_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127235968)))];
            tensor<fp16, [1, 1024, 1152]> linear_25_cast_fp16 = linear(bias = encoder_layers_4_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_k_proj_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127238336)))];
            tensor<fp16, [1152]> encoder_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129892608)))];
            tensor<fp16, [1, 1024, 1152]> linear_26_cast_fp16 = linear(bias = encoder_layers_4_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_v_proj_weight_to_fp16, x = hidden_states_25_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> var_344 = const()[name = tensor<string, []>("op_344"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_345_cast_fp16 = reshape(shape = var_344, x = linear_24_cast_fp16)[name = tensor<string, []>("op_345_cast_fp16")];
            tensor<int32, [4]> var_347 = const()[name = tensor<string, []>("op_347"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_348_cast_fp16 = reshape(shape = var_347, x = linear_25_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")];
            tensor<int32, [4]> var_350 = const()[name = tensor<string, []>("op_350"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_351_cast_fp16 = reshape(shape = var_350, x = linear_26_cast_fp16)[name = tensor<string, []>("op_351_cast_fp16")];
            tensor<int32, [4]> value_states_19_perm_0 = const()[name = tensor<string, []>("value_states_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_354_transpose_x_0 = const()[name = tensor<string, []>("op_354_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_354_transpose_y_0 = const()[name = tensor<string, []>("op_354_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_90 = transpose(perm = transpose_90_perm_0, x = var_348_cast_fp16)[name = tensor<string, []>("transpose_224")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_89 = transpose(perm = transpose_89_perm_0, x = var_345_cast_fp16)[name = tensor<string, []>("transpose_225")];
            tensor<fp16, [1, 16, 1024, 1024]> var_354_cast_fp16 = matmul(transpose_x = var_354_transpose_x_0, transpose_y = var_354_transpose_y_0, x = transpose_89, y = transpose_90)[name = tensor<string, []>("op_354_cast_fp16")];
            tensor<fp16, []> var_355_to_fp16 = const()[name = tensor<string, []>("op_355_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_17_cast_fp16 = mul(x = var_354_cast_fp16, y = var_355_to_fp16)[name = tensor<string, []>("attn_weights_17_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_357_cast_fp16 = softmax(axis = var_11, x = attn_weights_17_cast_fp16)[name = tensor<string, []>("op_357_cast_fp16")];
            tensor<bool, []> attn_output_17_transpose_x_0 = const()[name = tensor<string, []>("attn_output_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_17_transpose_y_0 = const()[name = tensor<string, []>("attn_output_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_19_cast_fp16 = transpose(perm = value_states_19_perm_0, x = var_351_cast_fp16)[name = tensor<string, []>("transpose_226")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_17_cast_fp16 = matmul(transpose_x = attn_output_17_transpose_x_0, transpose_y = attn_output_17_transpose_y_0, x = var_357_cast_fp16, y = value_states_19_cast_fp16)[name = tensor<string, []>("attn_output_17_cast_fp16")];
            tensor<int32, [4]> var_361_perm_0 = const()[name = tensor<string, []>("op_361_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_363 = const()[name = tensor<string, []>("op_363"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_361_cast_fp16 = transpose(perm = var_361_perm_0, x = attn_output_17_cast_fp16)[name = tensor<string, []>("transpose_223")];
            tensor<fp16, [1, 1024, 1152]> input_59_cast_fp16 = reshape(shape = var_363, x = var_361_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129894976)))];
            tensor<fp16, [1152]> encoder_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132549248)))];
            tensor<fp16, [1, 1024, 1152]> linear_27_cast_fp16 = linear(bias = encoder_layers_4_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_4_self_attn_out_proj_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_61_cast_fp16 = add(x = input_55_cast_fp16, y = linear_27_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")];
            tensor<int32, [1]> input_63_axes_0 = const()[name = tensor<string, []>("input_63_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_4_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132551616)))];
            tensor<fp16, [1152]> encoder_layers_4_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132553984)))];
            tensor<fp16, [1, 1024, 1152]> input_63_cast_fp16 = layer_norm(axes = input_63_axes_0, beta = encoder_layers_4_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_layer_norm2_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_4_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132556352)))];
            tensor<fp16, [4304]> encoder_layers_4_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142472832)))];
            tensor<fp16, [1, 1024, 4304]> linear_28_cast_fp16 = linear(bias = encoder_layers_4_mlp_fc1_bias_to_fp16, weight = encoder_layers_4_mlp_fc1_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<string, []> input_67_mode_0 = const()[name = tensor<string, []>("input_67_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_67_cast_fp16 = gelu(mode = input_67_mode_0, x = linear_28_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_4_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142481536)))];
            tensor<fp16, [1152]> encoder_layers_4_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152398016)))];
            tensor<fp16, [1, 1024, 1152]> linear_29_cast_fp16 = linear(bias = encoder_layers_4_mlp_fc2_bias_to_fp16, weight = encoder_layers_4_mlp_fc2_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_69_cast_fp16 = add(x = input_61_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
            tensor<int32, [1]> hidden_states_31_axes_0 = const()[name = tensor<string, []>("hidden_states_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_5_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152400384)))];
            tensor<fp16, [1152]> encoder_layers_5_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152402752)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_31_cast_fp16 = layer_norm(axes = hidden_states_31_axes_0, beta = encoder_layers_5_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_layer_norm1_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("hidden_states_31_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152405120)))];
            tensor<fp16, [1152]> encoder_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155059392)))];
            tensor<fp16, [1, 1024, 1152]> linear_30_cast_fp16 = linear(bias = encoder_layers_5_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_q_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155061760)))];
            tensor<fp16, [1152]> encoder_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157716032)))];
            tensor<fp16, [1, 1024, 1152]> linear_31_cast_fp16 = linear(bias = encoder_layers_5_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_k_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157718400)))];
            tensor<fp16, [1152]> encoder_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160372672)))];
            tensor<fp16, [1, 1024, 1152]> linear_32_cast_fp16 = linear(bias = encoder_layers_5_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_v_proj_weight_to_fp16, x = hidden_states_31_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_406 = const()[name = tensor<string, []>("op_406"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_407_cast_fp16 = reshape(shape = var_406, x = linear_30_cast_fp16)[name = tensor<string, []>("op_407_cast_fp16")];
            tensor<int32, [4]> var_409 = const()[name = tensor<string, []>("op_409"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_410_cast_fp16 = reshape(shape = var_409, x = linear_31_cast_fp16)[name = tensor<string, []>("op_410_cast_fp16")];
            tensor<int32, [4]> var_412 = const()[name = tensor<string, []>("op_412"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_413_cast_fp16 = reshape(shape = var_412, x = linear_32_cast_fp16)[name = tensor<string, []>("op_413_cast_fp16")];
            tensor<int32, [4]> value_states_23_perm_0 = const()[name = tensor<string, []>("value_states_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_416_transpose_x_0 = const()[name = tensor<string, []>("op_416_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_416_transpose_y_0 = const()[name = tensor<string, []>("op_416_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_92 = transpose(perm = transpose_92_perm_0, x = var_410_cast_fp16)[name = tensor<string, []>("transpose_220")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_91 = transpose(perm = transpose_91_perm_0, x = var_407_cast_fp16)[name = tensor<string, []>("transpose_221")];
            tensor<fp16, [1, 16, 1024, 1024]> var_416_cast_fp16 = matmul(transpose_x = var_416_transpose_x_0, transpose_y = var_416_transpose_y_0, x = transpose_91, y = transpose_92)[name = tensor<string, []>("op_416_cast_fp16")];
            tensor<fp16, []> var_417_to_fp16 = const()[name = tensor<string, []>("op_417_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_21_cast_fp16 = mul(x = var_416_cast_fp16, y = var_417_to_fp16)[name = tensor<string, []>("attn_weights_21_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_419_cast_fp16 = softmax(axis = var_11, x = attn_weights_21_cast_fp16)[name = tensor<string, []>("op_419_cast_fp16")];
            tensor<bool, []> attn_output_21_transpose_x_0 = const()[name = tensor<string, []>("attn_output_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_21_transpose_y_0 = const()[name = tensor<string, []>("attn_output_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_23_cast_fp16 = transpose(perm = value_states_23_perm_0, x = var_413_cast_fp16)[name = tensor<string, []>("transpose_222")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_21_cast_fp16 = matmul(transpose_x = attn_output_21_transpose_x_0, transpose_y = attn_output_21_transpose_y_0, x = var_419_cast_fp16, y = value_states_23_cast_fp16)[name = tensor<string, []>("attn_output_21_cast_fp16")];
            tensor<int32, [4]> var_423_perm_0 = const()[name = tensor<string, []>("op_423_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_425 = const()[name = tensor<string, []>("op_425"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_423_cast_fp16 = transpose(perm = var_423_perm_0, x = attn_output_21_cast_fp16)[name = tensor<string, []>("transpose_219")];
            tensor<fp16, [1, 1024, 1152]> input_73_cast_fp16 = reshape(shape = var_425, x = var_423_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160375040)))];
            tensor<fp16, [1152]> encoder_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163029312)))];
            tensor<fp16, [1, 1024, 1152]> linear_33_cast_fp16 = linear(bias = encoder_layers_5_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_5_self_attn_out_proj_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_75_cast_fp16 = add(x = input_69_cast_fp16, y = linear_33_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")];
            tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_5_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163031680)))];
            tensor<fp16, [1152]> encoder_layers_5_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163034048)))];
            tensor<fp16, [1, 1024, 1152]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = encoder_layers_5_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_layer_norm2_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_5_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163036416)))];
            tensor<fp16, [4304]> encoder_layers_5_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172952896)))];
            tensor<fp16, [1, 1024, 4304]> linear_34_cast_fp16 = linear(bias = encoder_layers_5_mlp_fc1_bias_to_fp16, weight = encoder_layers_5_mlp_fc1_weight_to_fp16, x = input_77_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<string, []> input_81_mode_0 = const()[name = tensor<string, []>("input_81_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_81_cast_fp16 = gelu(mode = input_81_mode_0, x = linear_34_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_5_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172961600)))];
            tensor<fp16, [1152]> encoder_layers_5_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(182878080)))];
            tensor<fp16, [1, 1024, 1152]> linear_35_cast_fp16 = linear(bias = encoder_layers_5_mlp_fc2_bias_to_fp16, weight = encoder_layers_5_mlp_fc2_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_83_cast_fp16 = add(x = input_75_cast_fp16, y = linear_35_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
            tensor<int32, [1]> hidden_states_37_axes_0 = const()[name = tensor<string, []>("hidden_states_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_6_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(182880448)))];
            tensor<fp16, [1152]> encoder_layers_6_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(182882816)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_37_cast_fp16 = layer_norm(axes = hidden_states_37_axes_0, beta = encoder_layers_6_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_layer_norm1_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("hidden_states_37_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(182885184)))];
            tensor<fp16, [1152]> encoder_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185539456)))];
            tensor<fp16, [1, 1024, 1152]> linear_36_cast_fp16 = linear(bias = encoder_layers_6_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_q_proj_weight_to_fp16, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185541824)))];
            tensor<fp16, [1152]> encoder_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188196096)))];
            tensor<fp16, [1, 1024, 1152]> linear_37_cast_fp16 = linear(bias = encoder_layers_6_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_k_proj_weight_to_fp16, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188198464)))];
            tensor<fp16, [1152]> encoder_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190852736)))];
            tensor<fp16, [1, 1024, 1152]> linear_38_cast_fp16 = linear(bias = encoder_layers_6_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_v_proj_weight_to_fp16, x = hidden_states_37_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<int32, [4]> var_468 = const()[name = tensor<string, []>("op_468"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_469_cast_fp16 = reshape(shape = var_468, x = linear_36_cast_fp16)[name = tensor<string, []>("op_469_cast_fp16")];
            tensor<int32, [4]> var_471 = const()[name = tensor<string, []>("op_471"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_472_cast_fp16 = reshape(shape = var_471, x = linear_37_cast_fp16)[name = tensor<string, []>("op_472_cast_fp16")];
            tensor<int32, [4]> var_474 = const()[name = tensor<string, []>("op_474"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_475_cast_fp16 = reshape(shape = var_474, x = linear_38_cast_fp16)[name = tensor<string, []>("op_475_cast_fp16")];
            tensor<int32, [4]> value_states_27_perm_0 = const()[name = tensor<string, []>("value_states_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_478_transpose_x_0 = const()[name = tensor<string, []>("op_478_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_478_transpose_y_0 = const()[name = tensor<string, []>("op_478_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_94 = transpose(perm = transpose_94_perm_0, x = var_472_cast_fp16)[name = tensor<string, []>("transpose_216")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_93 = transpose(perm = transpose_93_perm_0, x = var_469_cast_fp16)[name = tensor<string, []>("transpose_217")];
            tensor<fp16, [1, 16, 1024, 1024]> var_478_cast_fp16 = matmul(transpose_x = var_478_transpose_x_0, transpose_y = var_478_transpose_y_0, x = transpose_93, y = transpose_94)[name = tensor<string, []>("op_478_cast_fp16")];
            tensor<fp16, []> var_479_to_fp16 = const()[name = tensor<string, []>("op_479_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_25_cast_fp16 = mul(x = var_478_cast_fp16, y = var_479_to_fp16)[name = tensor<string, []>("attn_weights_25_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_481_cast_fp16 = softmax(axis = var_11, x = attn_weights_25_cast_fp16)[name = tensor<string, []>("op_481_cast_fp16")];
            tensor<bool, []> attn_output_25_transpose_x_0 = const()[name = tensor<string, []>("attn_output_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_25_transpose_y_0 = const()[name = tensor<string, []>("attn_output_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_27_cast_fp16 = transpose(perm = value_states_27_perm_0, x = var_475_cast_fp16)[name = tensor<string, []>("transpose_218")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_25_cast_fp16 = matmul(transpose_x = attn_output_25_transpose_x_0, transpose_y = attn_output_25_transpose_y_0, x = var_481_cast_fp16, y = value_states_27_cast_fp16)[name = tensor<string, []>("attn_output_25_cast_fp16")];
            tensor<int32, [4]> var_485_perm_0 = const()[name = tensor<string, []>("op_485_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_487 = const()[name = tensor<string, []>("op_487"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_485_cast_fp16 = transpose(perm = var_485_perm_0, x = attn_output_25_cast_fp16)[name = tensor<string, []>("transpose_215")];
            tensor<fp16, [1, 1024, 1152]> input_87_cast_fp16 = reshape(shape = var_487, x = var_485_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(190855104)))];
            tensor<fp16, [1152]> encoder_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193509376)))];
            tensor<fp16, [1, 1024, 1152]> linear_39_cast_fp16 = linear(bias = encoder_layers_6_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_6_self_attn_out_proj_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_89_cast_fp16 = add(x = input_83_cast_fp16, y = linear_39_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")];
            tensor<int32, [1]> input_91_axes_0 = const()[name = tensor<string, []>("input_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_6_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193511744)))];
            tensor<fp16, [1152]> encoder_layers_6_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193514112)))];
            tensor<fp16, [1, 1024, 1152]> input_91_cast_fp16 = layer_norm(axes = input_91_axes_0, beta = encoder_layers_6_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_layer_norm2_weight_to_fp16, x = input_89_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_6_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(193516480)))];
            tensor<fp16, [4304]> encoder_layers_6_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203432960)))];
            tensor<fp16, [1, 1024, 4304]> linear_40_cast_fp16 = linear(bias = encoder_layers_6_mlp_fc1_bias_to_fp16, weight = encoder_layers_6_mlp_fc1_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<string, []> input_95_mode_0 = const()[name = tensor<string, []>("input_95_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_95_cast_fp16 = gelu(mode = input_95_mode_0, x = linear_40_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_6_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203441664)))];
            tensor<fp16, [1152]> encoder_layers_6_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213358144)))];
            tensor<fp16, [1, 1024, 1152]> linear_41_cast_fp16 = linear(bias = encoder_layers_6_mlp_fc2_bias_to_fp16, weight = encoder_layers_6_mlp_fc2_weight_to_fp16, x = input_95_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_97_cast_fp16 = add(x = input_89_cast_fp16, y = linear_41_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<int32, [1]> hidden_states_43_axes_0 = const()[name = tensor<string, []>("hidden_states_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_7_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213360512)))];
            tensor<fp16, [1152]> encoder_layers_7_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213362880)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_43_cast_fp16 = layer_norm(axes = hidden_states_43_axes_0, beta = encoder_layers_7_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_layer_norm1_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("hidden_states_43_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213365248)))];
            tensor<fp16, [1152]> encoder_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216019520)))];
            tensor<fp16, [1, 1024, 1152]> linear_42_cast_fp16 = linear(bias = encoder_layers_7_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_q_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216021888)))];
            tensor<fp16, [1152]> encoder_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218676160)))];
            tensor<fp16, [1, 1024, 1152]> linear_43_cast_fp16 = linear(bias = encoder_layers_7_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_k_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(218678528)))];
            tensor<fp16, [1152]> encoder_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221332800)))];
            tensor<fp16, [1, 1024, 1152]> linear_44_cast_fp16 = linear(bias = encoder_layers_7_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_v_proj_weight_to_fp16, x = hidden_states_43_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> var_530 = const()[name = tensor<string, []>("op_530"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_531_cast_fp16 = reshape(shape = var_530, x = linear_42_cast_fp16)[name = tensor<string, []>("op_531_cast_fp16")];
            tensor<int32, [4]> var_533 = const()[name = tensor<string, []>("op_533"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_534_cast_fp16 = reshape(shape = var_533, x = linear_43_cast_fp16)[name = tensor<string, []>("op_534_cast_fp16")];
            tensor<int32, [4]> var_536 = const()[name = tensor<string, []>("op_536"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_537_cast_fp16 = reshape(shape = var_536, x = linear_44_cast_fp16)[name = tensor<string, []>("op_537_cast_fp16")];
            tensor<int32, [4]> value_states_31_perm_0 = const()[name = tensor<string, []>("value_states_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_540_transpose_x_0 = const()[name = tensor<string, []>("op_540_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_540_transpose_y_0 = const()[name = tensor<string, []>("op_540_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_534_cast_fp16)[name = tensor<string, []>("transpose_212")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_95 = transpose(perm = transpose_95_perm_0, x = var_531_cast_fp16)[name = tensor<string, []>("transpose_213")];
            tensor<fp16, [1, 16, 1024, 1024]> var_540_cast_fp16 = matmul(transpose_x = var_540_transpose_x_0, transpose_y = var_540_transpose_y_0, x = transpose_95, y = transpose_96)[name = tensor<string, []>("op_540_cast_fp16")];
            tensor<fp16, []> var_541_to_fp16 = const()[name = tensor<string, []>("op_541_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_29_cast_fp16 = mul(x = var_540_cast_fp16, y = var_541_to_fp16)[name = tensor<string, []>("attn_weights_29_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_543_cast_fp16 = softmax(axis = var_11, x = attn_weights_29_cast_fp16)[name = tensor<string, []>("op_543_cast_fp16")];
            tensor<bool, []> attn_output_29_transpose_x_0 = const()[name = tensor<string, []>("attn_output_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_29_transpose_y_0 = const()[name = tensor<string, []>("attn_output_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_31_cast_fp16 = transpose(perm = value_states_31_perm_0, x = var_537_cast_fp16)[name = tensor<string, []>("transpose_214")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_29_cast_fp16 = matmul(transpose_x = attn_output_29_transpose_x_0, transpose_y = attn_output_29_transpose_y_0, x = var_543_cast_fp16, y = value_states_31_cast_fp16)[name = tensor<string, []>("attn_output_29_cast_fp16")];
            tensor<int32, [4]> var_547_perm_0 = const()[name = tensor<string, []>("op_547_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_549 = const()[name = tensor<string, []>("op_549"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_547_cast_fp16 = transpose(perm = var_547_perm_0, x = attn_output_29_cast_fp16)[name = tensor<string, []>("transpose_211")];
            tensor<fp16, [1, 1024, 1152]> input_101_cast_fp16 = reshape(shape = var_549, x = var_547_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221335168)))];
            tensor<fp16, [1152]> encoder_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223989440)))];
            tensor<fp16, [1, 1024, 1152]> linear_45_cast_fp16 = linear(bias = encoder_layers_7_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_7_self_attn_out_proj_weight_to_fp16, x = input_101_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_103_cast_fp16 = add(x = input_97_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<int32, [1]> input_105_axes_0 = const()[name = tensor<string, []>("input_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_7_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223991808)))];
            tensor<fp16, [1152]> encoder_layers_7_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223994176)))];
            tensor<fp16, [1, 1024, 1152]> input_105_cast_fp16 = layer_norm(axes = input_105_axes_0, beta = encoder_layers_7_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_layer_norm2_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_7_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(223996544)))];
            tensor<fp16, [4304]> encoder_layers_7_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233913024)))];
            tensor<fp16, [1, 1024, 4304]> linear_46_cast_fp16 = linear(bias = encoder_layers_7_mlp_fc1_bias_to_fp16, weight = encoder_layers_7_mlp_fc1_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<string, []> input_109_mode_0 = const()[name = tensor<string, []>("input_109_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_109_cast_fp16 = gelu(mode = input_109_mode_0, x = linear_46_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_7_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233921728)))];
            tensor<fp16, [1152]> encoder_layers_7_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243838208)))];
            tensor<fp16, [1, 1024, 1152]> linear_47_cast_fp16 = linear(bias = encoder_layers_7_mlp_fc2_bias_to_fp16, weight = encoder_layers_7_mlp_fc2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_111_cast_fp16 = add(x = input_103_cast_fp16, y = linear_47_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")];
            tensor<int32, [1]> hidden_states_49_axes_0 = const()[name = tensor<string, []>("hidden_states_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_8_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243840576)))];
            tensor<fp16, [1152]> encoder_layers_8_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243842944)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_49_cast_fp16 = layer_norm(axes = hidden_states_49_axes_0, beta = encoder_layers_8_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_layer_norm1_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("hidden_states_49_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(243845312)))];
            tensor<fp16, [1152]> encoder_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246499584)))];
            tensor<fp16, [1, 1024, 1152]> linear_48_cast_fp16 = linear(bias = encoder_layers_8_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_q_proj_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246501952)))];
            tensor<fp16, [1152]> encoder_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249156224)))];
            tensor<fp16, [1, 1024, 1152]> linear_49_cast_fp16 = linear(bias = encoder_layers_8_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_k_proj_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249158592)))];
            tensor<fp16, [1152]> encoder_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251812864)))];
            tensor<fp16, [1, 1024, 1152]> linear_50_cast_fp16 = linear(bias = encoder_layers_8_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_v_proj_weight_to_fp16, x = hidden_states_49_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_592 = const()[name = tensor<string, []>("op_592"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_593_cast_fp16 = reshape(shape = var_592, x = linear_48_cast_fp16)[name = tensor<string, []>("op_593_cast_fp16")];
            tensor<int32, [4]> var_595 = const()[name = tensor<string, []>("op_595"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_596_cast_fp16 = reshape(shape = var_595, x = linear_49_cast_fp16)[name = tensor<string, []>("op_596_cast_fp16")];
            tensor<int32, [4]> var_598 = const()[name = tensor<string, []>("op_598"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_599_cast_fp16 = reshape(shape = var_598, x = linear_50_cast_fp16)[name = tensor<string, []>("op_599_cast_fp16")];
            tensor<int32, [4]> value_states_35_perm_0 = const()[name = tensor<string, []>("value_states_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_602_transpose_x_0 = const()[name = tensor<string, []>("op_602_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_602_transpose_y_0 = const()[name = tensor<string, []>("op_602_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_97_perm_0 = const()[name = tensor<string, []>("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_98_perm_0 = const()[name = tensor<string, []>("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_596_cast_fp16)[name = tensor<string, []>("transpose_208")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_97 = transpose(perm = transpose_97_perm_0, x = var_593_cast_fp16)[name = tensor<string, []>("transpose_209")];
            tensor<fp16, [1, 16, 1024, 1024]> var_602_cast_fp16 = matmul(transpose_x = var_602_transpose_x_0, transpose_y = var_602_transpose_y_0, x = transpose_97, y = transpose_98)[name = tensor<string, []>("op_602_cast_fp16")];
            tensor<fp16, []> var_603_to_fp16 = const()[name = tensor<string, []>("op_603_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_33_cast_fp16 = mul(x = var_602_cast_fp16, y = var_603_to_fp16)[name = tensor<string, []>("attn_weights_33_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_605_cast_fp16 = softmax(axis = var_11, x = attn_weights_33_cast_fp16)[name = tensor<string, []>("op_605_cast_fp16")];
            tensor<bool, []> attn_output_33_transpose_x_0 = const()[name = tensor<string, []>("attn_output_33_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_33_transpose_y_0 = const()[name = tensor<string, []>("attn_output_33_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_35_cast_fp16 = transpose(perm = value_states_35_perm_0, x = var_599_cast_fp16)[name = tensor<string, []>("transpose_210")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_33_cast_fp16 = matmul(transpose_x = attn_output_33_transpose_x_0, transpose_y = attn_output_33_transpose_y_0, x = var_605_cast_fp16, y = value_states_35_cast_fp16)[name = tensor<string, []>("attn_output_33_cast_fp16")];
            tensor<int32, [4]> var_609_perm_0 = const()[name = tensor<string, []>("op_609_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_611 = const()[name = tensor<string, []>("op_611"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_609_cast_fp16 = transpose(perm = var_609_perm_0, x = attn_output_33_cast_fp16)[name = tensor<string, []>("transpose_207")];
            tensor<fp16, [1, 1024, 1152]> input_115_cast_fp16 = reshape(shape = var_611, x = var_609_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(251815232)))];
            tensor<fp16, [1152]> encoder_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254469504)))];
            tensor<fp16, [1, 1024, 1152]> linear_51_cast_fp16 = linear(bias = encoder_layers_8_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_8_self_attn_out_proj_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_117_cast_fp16 = add(x = input_111_cast_fp16, y = linear_51_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")];
            tensor<int32, [1]> input_119_axes_0 = const()[name = tensor<string, []>("input_119_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_8_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254471872)))];
            tensor<fp16, [1152]> encoder_layers_8_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254474240)))];
            tensor<fp16, [1, 1024, 1152]> input_119_cast_fp16 = layer_norm(axes = input_119_axes_0, beta = encoder_layers_8_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_layer_norm2_weight_to_fp16, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_8_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(254476608)))];
            tensor<fp16, [4304]> encoder_layers_8_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264393088)))];
            tensor<fp16, [1, 1024, 4304]> linear_52_cast_fp16 = linear(bias = encoder_layers_8_mlp_fc1_bias_to_fp16, weight = encoder_layers_8_mlp_fc1_weight_to_fp16, x = input_119_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<string, []> input_123_mode_0 = const()[name = tensor<string, []>("input_123_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_123_cast_fp16 = gelu(mode = input_123_mode_0, x = linear_52_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_8_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(264401792)))];
            tensor<fp16, [1152]> encoder_layers_8_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274318272)))];
            tensor<fp16, [1, 1024, 1152]> linear_53_cast_fp16 = linear(bias = encoder_layers_8_mlp_fc2_bias_to_fp16, weight = encoder_layers_8_mlp_fc2_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_125_cast_fp16 = add(x = input_117_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<int32, [1]> hidden_states_55_axes_0 = const()[name = tensor<string, []>("hidden_states_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_9_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274320640)))];
            tensor<fp16, [1152]> encoder_layers_9_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274323008)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_55_cast_fp16 = layer_norm(axes = hidden_states_55_axes_0, beta = encoder_layers_9_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_layer_norm1_weight_to_fp16, x = input_125_cast_fp16)[name = tensor<string, []>("hidden_states_55_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(274325376)))];
            tensor<fp16, [1152]> encoder_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276979648)))];
            tensor<fp16, [1, 1024, 1152]> linear_54_cast_fp16 = linear(bias = encoder_layers_9_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_q_proj_weight_to_fp16, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276982016)))];
            tensor<fp16, [1152]> encoder_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(279636288)))];
            tensor<fp16, [1, 1024, 1152]> linear_55_cast_fp16 = linear(bias = encoder_layers_9_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_k_proj_weight_to_fp16, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(279638656)))];
            tensor<fp16, [1152]> encoder_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282292928)))];
            tensor<fp16, [1, 1024, 1152]> linear_56_cast_fp16 = linear(bias = encoder_layers_9_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_v_proj_weight_to_fp16, x = hidden_states_55_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<int32, [4]> var_654 = const()[name = tensor<string, []>("op_654"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_655_cast_fp16 = reshape(shape = var_654, x = linear_54_cast_fp16)[name = tensor<string, []>("op_655_cast_fp16")];
            tensor<int32, [4]> var_657 = const()[name = tensor<string, []>("op_657"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_658_cast_fp16 = reshape(shape = var_657, x = linear_55_cast_fp16)[name = tensor<string, []>("op_658_cast_fp16")];
            tensor<int32, [4]> var_660 = const()[name = tensor<string, []>("op_660"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_661_cast_fp16 = reshape(shape = var_660, x = linear_56_cast_fp16)[name = tensor<string, []>("op_661_cast_fp16")];
            tensor<int32, [4]> value_states_39_perm_0 = const()[name = tensor<string, []>("value_states_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_664_transpose_x_0 = const()[name = tensor<string, []>("op_664_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_664_transpose_y_0 = const()[name = tensor<string, []>("op_664_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_99_perm_0 = const()[name = tensor<string, []>("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_100_perm_0 = const()[name = tensor<string, []>("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_658_cast_fp16)[name = tensor<string, []>("transpose_204")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_99 = transpose(perm = transpose_99_perm_0, x = var_655_cast_fp16)[name = tensor<string, []>("transpose_205")];
            tensor<fp16, [1, 16, 1024, 1024]> var_664_cast_fp16 = matmul(transpose_x = var_664_transpose_x_0, transpose_y = var_664_transpose_y_0, x = transpose_99, y = transpose_100)[name = tensor<string, []>("op_664_cast_fp16")];
            tensor<fp16, []> var_665_to_fp16 = const()[name = tensor<string, []>("op_665_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_37_cast_fp16 = mul(x = var_664_cast_fp16, y = var_665_to_fp16)[name = tensor<string, []>("attn_weights_37_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_667_cast_fp16 = softmax(axis = var_11, x = attn_weights_37_cast_fp16)[name = tensor<string, []>("op_667_cast_fp16")];
            tensor<bool, []> attn_output_37_transpose_x_0 = const()[name = tensor<string, []>("attn_output_37_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_37_transpose_y_0 = const()[name = tensor<string, []>("attn_output_37_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_39_cast_fp16 = transpose(perm = value_states_39_perm_0, x = var_661_cast_fp16)[name = tensor<string, []>("transpose_206")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_37_cast_fp16 = matmul(transpose_x = attn_output_37_transpose_x_0, transpose_y = attn_output_37_transpose_y_0, x = var_667_cast_fp16, y = value_states_39_cast_fp16)[name = tensor<string, []>("attn_output_37_cast_fp16")];
            tensor<int32, [4]> var_671_perm_0 = const()[name = tensor<string, []>("op_671_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_673 = const()[name = tensor<string, []>("op_673"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_671_cast_fp16 = transpose(perm = var_671_perm_0, x = attn_output_37_cast_fp16)[name = tensor<string, []>("transpose_203")];
            tensor<fp16, [1, 1024, 1152]> input_129_cast_fp16 = reshape(shape = var_673, x = var_671_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282295296)))];
            tensor<fp16, [1152]> encoder_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284949568)))];
            tensor<fp16, [1, 1024, 1152]> linear_57_cast_fp16 = linear(bias = encoder_layers_9_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_9_self_attn_out_proj_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_131_cast_fp16 = add(x = input_125_cast_fp16, y = linear_57_cast_fp16)[name = tensor<string, []>("input_131_cast_fp16")];
            tensor<int32, [1]> input_133_axes_0 = const()[name = tensor<string, []>("input_133_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_9_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284951936)))];
            tensor<fp16, [1152]> encoder_layers_9_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284954304)))];
            tensor<fp16, [1, 1024, 1152]> input_133_cast_fp16 = layer_norm(axes = input_133_axes_0, beta = encoder_layers_9_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_layer_norm2_weight_to_fp16, x = input_131_cast_fp16)[name = tensor<string, []>("input_133_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_9_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284956672)))];
            tensor<fp16, [4304]> encoder_layers_9_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(294873152)))];
            tensor<fp16, [1, 1024, 4304]> linear_58_cast_fp16 = linear(bias = encoder_layers_9_mlp_fc1_bias_to_fp16, weight = encoder_layers_9_mlp_fc1_weight_to_fp16, x = input_133_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<string, []> input_137_mode_0 = const()[name = tensor<string, []>("input_137_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_137_cast_fp16 = gelu(mode = input_137_mode_0, x = linear_58_cast_fp16)[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_9_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(294881856)))];
            tensor<fp16, [1152]> encoder_layers_9_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304798336)))];
            tensor<fp16, [1, 1024, 1152]> linear_59_cast_fp16 = linear(bias = encoder_layers_9_mlp_fc2_bias_to_fp16, weight = encoder_layers_9_mlp_fc2_weight_to_fp16, x = input_137_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_139_cast_fp16 = add(x = input_131_cast_fp16, y = linear_59_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<int32, [1]> hidden_states_61_axes_0 = const()[name = tensor<string, []>("hidden_states_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_10_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304800704)))];
            tensor<fp16, [1152]> encoder_layers_10_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304803072)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_61_cast_fp16 = layer_norm(axes = hidden_states_61_axes_0, beta = encoder_layers_10_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_layer_norm1_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("hidden_states_61_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(304805440)))];
            tensor<fp16, [1152]> encoder_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307459712)))];
            tensor<fp16, [1, 1024, 1152]> linear_60_cast_fp16 = linear(bias = encoder_layers_10_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_q_proj_weight_to_fp16, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(307462080)))];
            tensor<fp16, [1152]> encoder_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310116352)))];
            tensor<fp16, [1, 1024, 1152]> linear_61_cast_fp16 = linear(bias = encoder_layers_10_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_k_proj_weight_to_fp16, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(310118720)))];
            tensor<fp16, [1152]> encoder_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312772992)))];
            tensor<fp16, [1, 1024, 1152]> linear_62_cast_fp16 = linear(bias = encoder_layers_10_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_v_proj_weight_to_fp16, x = hidden_states_61_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<int32, [4]> var_716 = const()[name = tensor<string, []>("op_716"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_717_cast_fp16 = reshape(shape = var_716, x = linear_60_cast_fp16)[name = tensor<string, []>("op_717_cast_fp16")];
            tensor<int32, [4]> var_719 = const()[name = tensor<string, []>("op_719"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_720_cast_fp16 = reshape(shape = var_719, x = linear_61_cast_fp16)[name = tensor<string, []>("op_720_cast_fp16")];
            tensor<int32, [4]> var_722 = const()[name = tensor<string, []>("op_722"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_723_cast_fp16 = reshape(shape = var_722, x = linear_62_cast_fp16)[name = tensor<string, []>("op_723_cast_fp16")];
            tensor<int32, [4]> value_states_43_perm_0 = const()[name = tensor<string, []>("value_states_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_726_transpose_x_0 = const()[name = tensor<string, []>("op_726_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_726_transpose_y_0 = const()[name = tensor<string, []>("op_726_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_102_perm_0 = const()[name = tensor<string, []>("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_102 = transpose(perm = transpose_102_perm_0, x = var_720_cast_fp16)[name = tensor<string, []>("transpose_200")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_101 = transpose(perm = transpose_101_perm_0, x = var_717_cast_fp16)[name = tensor<string, []>("transpose_201")];
            tensor<fp16, [1, 16, 1024, 1024]> var_726_cast_fp16 = matmul(transpose_x = var_726_transpose_x_0, transpose_y = var_726_transpose_y_0, x = transpose_101, y = transpose_102)[name = tensor<string, []>("op_726_cast_fp16")];
            tensor<fp16, []> var_727_to_fp16 = const()[name = tensor<string, []>("op_727_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_41_cast_fp16 = mul(x = var_726_cast_fp16, y = var_727_to_fp16)[name = tensor<string, []>("attn_weights_41_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_729_cast_fp16 = softmax(axis = var_11, x = attn_weights_41_cast_fp16)[name = tensor<string, []>("op_729_cast_fp16")];
            tensor<bool, []> attn_output_41_transpose_x_0 = const()[name = tensor<string, []>("attn_output_41_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_41_transpose_y_0 = const()[name = tensor<string, []>("attn_output_41_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_43_cast_fp16 = transpose(perm = value_states_43_perm_0, x = var_723_cast_fp16)[name = tensor<string, []>("transpose_202")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_41_cast_fp16 = matmul(transpose_x = attn_output_41_transpose_x_0, transpose_y = attn_output_41_transpose_y_0, x = var_729_cast_fp16, y = value_states_43_cast_fp16)[name = tensor<string, []>("attn_output_41_cast_fp16")];
            tensor<int32, [4]> var_733_perm_0 = const()[name = tensor<string, []>("op_733_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_735 = const()[name = tensor<string, []>("op_735"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_733_cast_fp16 = transpose(perm = var_733_perm_0, x = attn_output_41_cast_fp16)[name = tensor<string, []>("transpose_199")];
            tensor<fp16, [1, 1024, 1152]> input_143_cast_fp16 = reshape(shape = var_735, x = var_733_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(312775360)))];
            tensor<fp16, [1152]> encoder_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315429632)))];
            tensor<fp16, [1, 1024, 1152]> linear_63_cast_fp16 = linear(bias = encoder_layers_10_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_10_self_attn_out_proj_weight_to_fp16, x = input_143_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_145_cast_fp16 = add(x = input_139_cast_fp16, y = linear_63_cast_fp16)[name = tensor<string, []>("input_145_cast_fp16")];
            tensor<int32, [1]> input_147_axes_0 = const()[name = tensor<string, []>("input_147_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_10_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315432000)))];
            tensor<fp16, [1152]> encoder_layers_10_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315434368)))];
            tensor<fp16, [1, 1024, 1152]> input_147_cast_fp16 = layer_norm(axes = input_147_axes_0, beta = encoder_layers_10_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_layer_norm2_weight_to_fp16, x = input_145_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_10_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(315436736)))];
            tensor<fp16, [4304]> encoder_layers_10_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325353216)))];
            tensor<fp16, [1, 1024, 4304]> linear_64_cast_fp16 = linear(bias = encoder_layers_10_mlp_fc1_bias_to_fp16, weight = encoder_layers_10_mlp_fc1_weight_to_fp16, x = input_147_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<string, []> input_151_mode_0 = const()[name = tensor<string, []>("input_151_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_151_cast_fp16 = gelu(mode = input_151_mode_0, x = linear_64_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_10_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325361920)))];
            tensor<fp16, [1152]> encoder_layers_10_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335278400)))];
            tensor<fp16, [1, 1024, 1152]> linear_65_cast_fp16 = linear(bias = encoder_layers_10_mlp_fc2_bias_to_fp16, weight = encoder_layers_10_mlp_fc2_weight_to_fp16, x = input_151_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_153_cast_fp16 = add(x = input_145_cast_fp16, y = linear_65_cast_fp16)[name = tensor<string, []>("input_153_cast_fp16")];
            tensor<int32, [1]> hidden_states_67_axes_0 = const()[name = tensor<string, []>("hidden_states_67_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_11_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335280768)))];
            tensor<fp16, [1152]> encoder_layers_11_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335283136)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_67_cast_fp16 = layer_norm(axes = hidden_states_67_axes_0, beta = encoder_layers_11_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_layer_norm1_weight_to_fp16, x = input_153_cast_fp16)[name = tensor<string, []>("hidden_states_67_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335285504)))];
            tensor<fp16, [1152]> encoder_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337939776)))];
            tensor<fp16, [1, 1024, 1152]> linear_66_cast_fp16 = linear(bias = encoder_layers_11_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_q_proj_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(337942144)))];
            tensor<fp16, [1152]> encoder_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340596416)))];
            tensor<fp16, [1, 1024, 1152]> linear_67_cast_fp16 = linear(bias = encoder_layers_11_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_k_proj_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(340598784)))];
            tensor<fp16, [1152]> encoder_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343253056)))];
            tensor<fp16, [1, 1024, 1152]> linear_68_cast_fp16 = linear(bias = encoder_layers_11_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_v_proj_weight_to_fp16, x = hidden_states_67_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_778 = const()[name = tensor<string, []>("op_778"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_779_cast_fp16 = reshape(shape = var_778, x = linear_66_cast_fp16)[name = tensor<string, []>("op_779_cast_fp16")];
            tensor<int32, [4]> var_781 = const()[name = tensor<string, []>("op_781"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_782_cast_fp16 = reshape(shape = var_781, x = linear_67_cast_fp16)[name = tensor<string, []>("op_782_cast_fp16")];
            tensor<int32, [4]> var_784 = const()[name = tensor<string, []>("op_784"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_785_cast_fp16 = reshape(shape = var_784, x = linear_68_cast_fp16)[name = tensor<string, []>("op_785_cast_fp16")];
            tensor<int32, [4]> value_states_47_perm_0 = const()[name = tensor<string, []>("value_states_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_788_transpose_x_0 = const()[name = tensor<string, []>("op_788_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_788_transpose_y_0 = const()[name = tensor<string, []>("op_788_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_103_perm_0 = const()[name = tensor<string, []>("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_104_perm_0 = const()[name = tensor<string, []>("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_104 = transpose(perm = transpose_104_perm_0, x = var_782_cast_fp16)[name = tensor<string, []>("transpose_196")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_103 = transpose(perm = transpose_103_perm_0, x = var_779_cast_fp16)[name = tensor<string, []>("transpose_197")];
            tensor<fp16, [1, 16, 1024, 1024]> var_788_cast_fp16 = matmul(transpose_x = var_788_transpose_x_0, transpose_y = var_788_transpose_y_0, x = transpose_103, y = transpose_104)[name = tensor<string, []>("op_788_cast_fp16")];
            tensor<fp16, []> var_789_to_fp16 = const()[name = tensor<string, []>("op_789_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_45_cast_fp16 = mul(x = var_788_cast_fp16, y = var_789_to_fp16)[name = tensor<string, []>("attn_weights_45_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_791_cast_fp16 = softmax(axis = var_11, x = attn_weights_45_cast_fp16)[name = tensor<string, []>("op_791_cast_fp16")];
            tensor<bool, []> attn_output_45_transpose_x_0 = const()[name = tensor<string, []>("attn_output_45_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_45_transpose_y_0 = const()[name = tensor<string, []>("attn_output_45_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_47_cast_fp16 = transpose(perm = value_states_47_perm_0, x = var_785_cast_fp16)[name = tensor<string, []>("transpose_198")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_45_cast_fp16 = matmul(transpose_x = attn_output_45_transpose_x_0, transpose_y = attn_output_45_transpose_y_0, x = var_791_cast_fp16, y = value_states_47_cast_fp16)[name = tensor<string, []>("attn_output_45_cast_fp16")];
            tensor<int32, [4]> var_795_perm_0 = const()[name = tensor<string, []>("op_795_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_797 = const()[name = tensor<string, []>("op_797"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_795_cast_fp16 = transpose(perm = var_795_perm_0, x = attn_output_45_cast_fp16)[name = tensor<string, []>("transpose_195")];
            tensor<fp16, [1, 1024, 1152]> input_157_cast_fp16 = reshape(shape = var_797, x = var_795_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(343255424)))];
            tensor<fp16, [1152]> encoder_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345909696)))];
            tensor<fp16, [1, 1024, 1152]> linear_69_cast_fp16 = linear(bias = encoder_layers_11_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_11_self_attn_out_proj_weight_to_fp16, x = input_157_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_159_cast_fp16 = add(x = input_153_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("input_159_cast_fp16")];
            tensor<int32, [1]> input_161_axes_0 = const()[name = tensor<string, []>("input_161_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_11_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345912064)))];
            tensor<fp16, [1152]> encoder_layers_11_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345914432)))];
            tensor<fp16, [1, 1024, 1152]> input_161_cast_fp16 = layer_norm(axes = input_161_axes_0, beta = encoder_layers_11_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_layer_norm2_weight_to_fp16, x = input_159_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_11_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(345916800)))];
            tensor<fp16, [4304]> encoder_layers_11_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355833280)))];
            tensor<fp16, [1, 1024, 4304]> linear_70_cast_fp16 = linear(bias = encoder_layers_11_mlp_fc1_bias_to_fp16, weight = encoder_layers_11_mlp_fc1_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<string, []> input_165_mode_0 = const()[name = tensor<string, []>("input_165_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_165_cast_fp16 = gelu(mode = input_165_mode_0, x = linear_70_cast_fp16)[name = tensor<string, []>("input_165_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_11_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355841984)))];
            tensor<fp16, [1152]> encoder_layers_11_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365758464)))];
            tensor<fp16, [1, 1024, 1152]> linear_71_cast_fp16 = linear(bias = encoder_layers_11_mlp_fc2_bias_to_fp16, weight = encoder_layers_11_mlp_fc2_weight_to_fp16, x = input_165_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_167_cast_fp16 = add(x = input_159_cast_fp16, y = linear_71_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")];
            tensor<int32, [1]> hidden_states_73_axes_0 = const()[name = tensor<string, []>("hidden_states_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_12_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365760832)))];
            tensor<fp16, [1152]> encoder_layers_12_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365763200)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_73_cast_fp16 = layer_norm(axes = hidden_states_73_axes_0, beta = encoder_layers_12_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_layer_norm1_weight_to_fp16, x = input_167_cast_fp16)[name = tensor<string, []>("hidden_states_73_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(365765568)))];
            tensor<fp16, [1152]> encoder_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368419840)))];
            tensor<fp16, [1, 1024, 1152]> linear_72_cast_fp16 = linear(bias = encoder_layers_12_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_q_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368422208)))];
            tensor<fp16, [1152]> encoder_layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371076480)))];
            tensor<fp16, [1, 1024, 1152]> linear_73_cast_fp16 = linear(bias = encoder_layers_12_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_k_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371078848)))];
            tensor<fp16, [1152]> encoder_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373733120)))];
            tensor<fp16, [1, 1024, 1152]> linear_74_cast_fp16 = linear(bias = encoder_layers_12_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_v_proj_weight_to_fp16, x = hidden_states_73_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<int32, [4]> var_840 = const()[name = tensor<string, []>("op_840"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_841_cast_fp16 = reshape(shape = var_840, x = linear_72_cast_fp16)[name = tensor<string, []>("op_841_cast_fp16")];
            tensor<int32, [4]> var_843 = const()[name = tensor<string, []>("op_843"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_844_cast_fp16 = reshape(shape = var_843, x = linear_73_cast_fp16)[name = tensor<string, []>("op_844_cast_fp16")];
            tensor<int32, [4]> var_846 = const()[name = tensor<string, []>("op_846"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_847_cast_fp16 = reshape(shape = var_846, x = linear_74_cast_fp16)[name = tensor<string, []>("op_847_cast_fp16")];
            tensor<int32, [4]> value_states_51_perm_0 = const()[name = tensor<string, []>("value_states_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_850_transpose_x_0 = const()[name = tensor<string, []>("op_850_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_850_transpose_y_0 = const()[name = tensor<string, []>("op_850_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_105_perm_0 = const()[name = tensor<string, []>("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_106_perm_0 = const()[name = tensor<string, []>("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_106 = transpose(perm = transpose_106_perm_0, x = var_844_cast_fp16)[name = tensor<string, []>("transpose_192")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_105 = transpose(perm = transpose_105_perm_0, x = var_841_cast_fp16)[name = tensor<string, []>("transpose_193")];
            tensor<fp16, [1, 16, 1024, 1024]> var_850_cast_fp16 = matmul(transpose_x = var_850_transpose_x_0, transpose_y = var_850_transpose_y_0, x = transpose_105, y = transpose_106)[name = tensor<string, []>("op_850_cast_fp16")];
            tensor<fp16, []> var_851_to_fp16 = const()[name = tensor<string, []>("op_851_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_49_cast_fp16 = mul(x = var_850_cast_fp16, y = var_851_to_fp16)[name = tensor<string, []>("attn_weights_49_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_853_cast_fp16 = softmax(axis = var_11, x = attn_weights_49_cast_fp16)[name = tensor<string, []>("op_853_cast_fp16")];
            tensor<bool, []> attn_output_49_transpose_x_0 = const()[name = tensor<string, []>("attn_output_49_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_49_transpose_y_0 = const()[name = tensor<string, []>("attn_output_49_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_51_cast_fp16 = transpose(perm = value_states_51_perm_0, x = var_847_cast_fp16)[name = tensor<string, []>("transpose_194")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_49_cast_fp16 = matmul(transpose_x = attn_output_49_transpose_x_0, transpose_y = attn_output_49_transpose_y_0, x = var_853_cast_fp16, y = value_states_51_cast_fp16)[name = tensor<string, []>("attn_output_49_cast_fp16")];
            tensor<int32, [4]> var_857_perm_0 = const()[name = tensor<string, []>("op_857_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_859 = const()[name = tensor<string, []>("op_859"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_857_cast_fp16 = transpose(perm = var_857_perm_0, x = attn_output_49_cast_fp16)[name = tensor<string, []>("transpose_191")];
            tensor<fp16, [1, 1024, 1152]> input_171_cast_fp16 = reshape(shape = var_859, x = var_857_cast_fp16)[name = tensor<string, []>("input_171_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_12_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373735488)))];
            tensor<fp16, [1152]> encoder_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(376389760)))];
            tensor<fp16, [1, 1024, 1152]> linear_75_cast_fp16 = linear(bias = encoder_layers_12_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_12_self_attn_out_proj_weight_to_fp16, x = input_171_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_173_cast_fp16 = add(x = input_167_cast_fp16, y = linear_75_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")];
            tensor<int32, [1]> input_175_axes_0 = const()[name = tensor<string, []>("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_12_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(376392128)))];
            tensor<fp16, [1152]> encoder_layers_12_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(376394496)))];
            tensor<fp16, [1, 1024, 1152]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = encoder_layers_12_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_layer_norm2_weight_to_fp16, x = input_173_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_12_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(376396864)))];
            tensor<fp16, [4304]> encoder_layers_12_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386313344)))];
            tensor<fp16, [1, 1024, 4304]> linear_76_cast_fp16 = linear(bias = encoder_layers_12_mlp_fc1_bias_to_fp16, weight = encoder_layers_12_mlp_fc1_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
            tensor<string, []> input_179_mode_0 = const()[name = tensor<string, []>("input_179_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_179_cast_fp16 = gelu(mode = input_179_mode_0, x = linear_76_cast_fp16)[name = tensor<string, []>("input_179_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_12_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(386322048)))];
            tensor<fp16, [1152]> encoder_layers_12_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396238528)))];
            tensor<fp16, [1, 1024, 1152]> linear_77_cast_fp16 = linear(bias = encoder_layers_12_mlp_fc2_bias_to_fp16, weight = encoder_layers_12_mlp_fc2_weight_to_fp16, x = input_179_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_181_cast_fp16 = add(x = input_173_cast_fp16, y = linear_77_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<int32, [1]> hidden_states_79_axes_0 = const()[name = tensor<string, []>("hidden_states_79_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_13_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396240896)))];
            tensor<fp16, [1152]> encoder_layers_13_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396243264)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_79_cast_fp16 = layer_norm(axes = hidden_states_79_axes_0, beta = encoder_layers_13_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_layer_norm1_weight_to_fp16, x = input_181_cast_fp16)[name = tensor<string, []>("hidden_states_79_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(396245632)))];
            tensor<fp16, [1152]> encoder_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(398899904)))];
            tensor<fp16, [1, 1024, 1152]> linear_78_cast_fp16 = linear(bias = encoder_layers_13_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_q_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(398902272)))];
            tensor<fp16, [1152]> encoder_layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(401556544)))];
            tensor<fp16, [1, 1024, 1152]> linear_79_cast_fp16 = linear(bias = encoder_layers_13_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_k_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(401558912)))];
            tensor<fp16, [1152]> encoder_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(404213184)))];
            tensor<fp16, [1, 1024, 1152]> linear_80_cast_fp16 = linear(bias = encoder_layers_13_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_v_proj_weight_to_fp16, x = hidden_states_79_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
            tensor<int32, [4]> var_902 = const()[name = tensor<string, []>("op_902"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_903_cast_fp16 = reshape(shape = var_902, x = linear_78_cast_fp16)[name = tensor<string, []>("op_903_cast_fp16")];
            tensor<int32, [4]> var_905 = const()[name = tensor<string, []>("op_905"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_906_cast_fp16 = reshape(shape = var_905, x = linear_79_cast_fp16)[name = tensor<string, []>("op_906_cast_fp16")];
            tensor<int32, [4]> var_908 = const()[name = tensor<string, []>("op_908"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_909_cast_fp16 = reshape(shape = var_908, x = linear_80_cast_fp16)[name = tensor<string, []>("op_909_cast_fp16")];
            tensor<int32, [4]> value_states_55_perm_0 = const()[name = tensor<string, []>("value_states_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_912_transpose_x_0 = const()[name = tensor<string, []>("op_912_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_912_transpose_y_0 = const()[name = tensor<string, []>("op_912_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_107_perm_0 = const()[name = tensor<string, []>("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_108_perm_0 = const()[name = tensor<string, []>("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_108 = transpose(perm = transpose_108_perm_0, x = var_906_cast_fp16)[name = tensor<string, []>("transpose_188")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_107 = transpose(perm = transpose_107_perm_0, x = var_903_cast_fp16)[name = tensor<string, []>("transpose_189")];
            tensor<fp16, [1, 16, 1024, 1024]> var_912_cast_fp16 = matmul(transpose_x = var_912_transpose_x_0, transpose_y = var_912_transpose_y_0, x = transpose_107, y = transpose_108)[name = tensor<string, []>("op_912_cast_fp16")];
            tensor<fp16, []> var_913_to_fp16 = const()[name = tensor<string, []>("op_913_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_53_cast_fp16 = mul(x = var_912_cast_fp16, y = var_913_to_fp16)[name = tensor<string, []>("attn_weights_53_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_915_cast_fp16 = softmax(axis = var_11, x = attn_weights_53_cast_fp16)[name = tensor<string, []>("op_915_cast_fp16")];
            tensor<bool, []> attn_output_53_transpose_x_0 = const()[name = tensor<string, []>("attn_output_53_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_53_transpose_y_0 = const()[name = tensor<string, []>("attn_output_53_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_55_cast_fp16 = transpose(perm = value_states_55_perm_0, x = var_909_cast_fp16)[name = tensor<string, []>("transpose_190")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_53_cast_fp16 = matmul(transpose_x = attn_output_53_transpose_x_0, transpose_y = attn_output_53_transpose_y_0, x = var_915_cast_fp16, y = value_states_55_cast_fp16)[name = tensor<string, []>("attn_output_53_cast_fp16")];
            tensor<int32, [4]> var_919_perm_0 = const()[name = tensor<string, []>("op_919_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_921 = const()[name = tensor<string, []>("op_921"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_919_cast_fp16 = transpose(perm = var_919_perm_0, x = attn_output_53_cast_fp16)[name = tensor<string, []>("transpose_187")];
            tensor<fp16, [1, 1024, 1152]> input_185_cast_fp16 = reshape(shape = var_921, x = var_919_cast_fp16)[name = tensor<string, []>("input_185_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_13_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(404215552)))];
            tensor<fp16, [1152]> encoder_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406869824)))];
            tensor<fp16, [1, 1024, 1152]> linear_81_cast_fp16 = linear(bias = encoder_layers_13_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_13_self_attn_out_proj_weight_to_fp16, x = input_185_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_187_cast_fp16 = add(x = input_181_cast_fp16, y = linear_81_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<int32, [1]> input_189_axes_0 = const()[name = tensor<string, []>("input_189_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_13_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406872192)))];
            tensor<fp16, [1152]> encoder_layers_13_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406874560)))];
            tensor<fp16, [1, 1024, 1152]> input_189_cast_fp16 = layer_norm(axes = input_189_axes_0, beta = encoder_layers_13_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_layer_norm2_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("input_189_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_13_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(406876928)))];
            tensor<fp16, [4304]> encoder_layers_13_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416793408)))];
            tensor<fp16, [1, 1024, 4304]> linear_82_cast_fp16 = linear(bias = encoder_layers_13_mlp_fc1_bias_to_fp16, weight = encoder_layers_13_mlp_fc1_weight_to_fp16, x = input_189_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")];
            tensor<string, []> input_193_mode_0 = const()[name = tensor<string, []>("input_193_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_193_cast_fp16 = gelu(mode = input_193_mode_0, x = linear_82_cast_fp16)[name = tensor<string, []>("input_193_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_13_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(416802112)))];
            tensor<fp16, [1152]> encoder_layers_13_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426718592)))];
            tensor<fp16, [1, 1024, 1152]> linear_83_cast_fp16 = linear(bias = encoder_layers_13_mlp_fc2_bias_to_fp16, weight = encoder_layers_13_mlp_fc2_weight_to_fp16, x = input_193_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_83_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
            tensor<int32, [1]> hidden_states_85_axes_0 = const()[name = tensor<string, []>("hidden_states_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_14_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426720960)))];
            tensor<fp16, [1152]> encoder_layers_14_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426723328)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_85_cast_fp16 = layer_norm(axes = hidden_states_85_axes_0, beta = encoder_layers_14_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_layer_norm1_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("hidden_states_85_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(426725696)))];
            tensor<fp16, [1152]> encoder_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429379968)))];
            tensor<fp16, [1, 1024, 1152]> linear_84_cast_fp16 = linear(bias = encoder_layers_14_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_q_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(429382336)))];
            tensor<fp16, [1152]> encoder_layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(432036608)))];
            tensor<fp16, [1, 1024, 1152]> linear_85_cast_fp16 = linear(bias = encoder_layers_14_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_k_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(432038976)))];
            tensor<fp16, [1152]> encoder_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434693248)))];
            tensor<fp16, [1, 1024, 1152]> linear_86_cast_fp16 = linear(bias = encoder_layers_14_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_v_proj_weight_to_fp16, x = hidden_states_85_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
            tensor<int32, [4]> var_964 = const()[name = tensor<string, []>("op_964"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_965_cast_fp16 = reshape(shape = var_964, x = linear_84_cast_fp16)[name = tensor<string, []>("op_965_cast_fp16")];
            tensor<int32, [4]> var_967 = const()[name = tensor<string, []>("op_967"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_968_cast_fp16 = reshape(shape = var_967, x = linear_85_cast_fp16)[name = tensor<string, []>("op_968_cast_fp16")];
            tensor<int32, [4]> var_970 = const()[name = tensor<string, []>("op_970"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_971_cast_fp16 = reshape(shape = var_970, x = linear_86_cast_fp16)[name = tensor<string, []>("op_971_cast_fp16")];
            tensor<int32, [4]> value_states_59_perm_0 = const()[name = tensor<string, []>("value_states_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_974_transpose_x_0 = const()[name = tensor<string, []>("op_974_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_974_transpose_y_0 = const()[name = tensor<string, []>("op_974_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_109_perm_0 = const()[name = tensor<string, []>("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_110_perm_0 = const()[name = tensor<string, []>("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_110 = transpose(perm = transpose_110_perm_0, x = var_968_cast_fp16)[name = tensor<string, []>("transpose_184")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_109 = transpose(perm = transpose_109_perm_0, x = var_965_cast_fp16)[name = tensor<string, []>("transpose_185")];
            tensor<fp16, [1, 16, 1024, 1024]> var_974_cast_fp16 = matmul(transpose_x = var_974_transpose_x_0, transpose_y = var_974_transpose_y_0, x = transpose_109, y = transpose_110)[name = tensor<string, []>("op_974_cast_fp16")];
            tensor<fp16, []> var_975_to_fp16 = const()[name = tensor<string, []>("op_975_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_57_cast_fp16 = mul(x = var_974_cast_fp16, y = var_975_to_fp16)[name = tensor<string, []>("attn_weights_57_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_977_cast_fp16 = softmax(axis = var_11, x = attn_weights_57_cast_fp16)[name = tensor<string, []>("op_977_cast_fp16")];
            tensor<bool, []> attn_output_57_transpose_x_0 = const()[name = tensor<string, []>("attn_output_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_57_transpose_y_0 = const()[name = tensor<string, []>("attn_output_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_59_cast_fp16 = transpose(perm = value_states_59_perm_0, x = var_971_cast_fp16)[name = tensor<string, []>("transpose_186")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_57_cast_fp16 = matmul(transpose_x = attn_output_57_transpose_x_0, transpose_y = attn_output_57_transpose_y_0, x = var_977_cast_fp16, y = value_states_59_cast_fp16)[name = tensor<string, []>("attn_output_57_cast_fp16")];
            tensor<int32, [4]> var_981_perm_0 = const()[name = tensor<string, []>("op_981_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_983 = const()[name = tensor<string, []>("op_983"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_981_cast_fp16 = transpose(perm = var_981_perm_0, x = attn_output_57_cast_fp16)[name = tensor<string, []>("transpose_183")];
            tensor<fp16, [1, 1024, 1152]> input_199_cast_fp16 = reshape(shape = var_983, x = var_981_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_14_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434695616)))];
            tensor<fp16, [1152]> encoder_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(437349888)))];
            tensor<fp16, [1, 1024, 1152]> linear_87_cast_fp16 = linear(bias = encoder_layers_14_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_14_self_attn_out_proj_weight_to_fp16, x = input_199_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_201_cast_fp16 = add(x = input_195_cast_fp16, y = linear_87_cast_fp16)[name = tensor<string, []>("input_201_cast_fp16")];
            tensor<int32, [1]> input_203_axes_0 = const()[name = tensor<string, []>("input_203_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_14_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(437352256)))];
            tensor<fp16, [1152]> encoder_layers_14_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(437354624)))];
            tensor<fp16, [1, 1024, 1152]> input_203_cast_fp16 = layer_norm(axes = input_203_axes_0, beta = encoder_layers_14_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_layer_norm2_weight_to_fp16, x = input_201_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_14_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(437356992)))];
            tensor<fp16, [4304]> encoder_layers_14_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447273472)))];
            tensor<fp16, [1, 1024, 4304]> linear_88_cast_fp16 = linear(bias = encoder_layers_14_mlp_fc1_bias_to_fp16, weight = encoder_layers_14_mlp_fc1_weight_to_fp16, x = input_203_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")];
            tensor<string, []> input_207_mode_0 = const()[name = tensor<string, []>("input_207_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_207_cast_fp16 = gelu(mode = input_207_mode_0, x = linear_88_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_14_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447282176)))];
            tensor<fp16, [1152]> encoder_layers_14_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457198656)))];
            tensor<fp16, [1, 1024, 1152]> linear_89_cast_fp16 = linear(bias = encoder_layers_14_mlp_fc2_bias_to_fp16, weight = encoder_layers_14_mlp_fc2_weight_to_fp16, x = input_207_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_209_cast_fp16 = add(x = input_201_cast_fp16, y = linear_89_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")];
            tensor<int32, [1]> hidden_states_91_axes_0 = const()[name = tensor<string, []>("hidden_states_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_15_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457201024)))];
            tensor<fp16, [1152]> encoder_layers_15_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457203392)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_91_cast_fp16 = layer_norm(axes = hidden_states_91_axes_0, beta = encoder_layers_15_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_layer_norm1_weight_to_fp16, x = input_209_cast_fp16)[name = tensor<string, []>("hidden_states_91_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457205760)))];
            tensor<fp16, [1152]> encoder_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(459860032)))];
            tensor<fp16, [1, 1024, 1152]> linear_90_cast_fp16 = linear(bias = encoder_layers_15_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_q_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(459862400)))];
            tensor<fp16, [1152]> encoder_layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(462516672)))];
            tensor<fp16, [1, 1024, 1152]> linear_91_cast_fp16 = linear(bias = encoder_layers_15_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_k_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(462519040)))];
            tensor<fp16, [1152]> encoder_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465173312)))];
            tensor<fp16, [1, 1024, 1152]> linear_92_cast_fp16 = linear(bias = encoder_layers_15_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_v_proj_weight_to_fp16, x = hidden_states_91_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
            tensor<int32, [4]> var_1026 = const()[name = tensor<string, []>("op_1026"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1027_cast_fp16 = reshape(shape = var_1026, x = linear_90_cast_fp16)[name = tensor<string, []>("op_1027_cast_fp16")];
            tensor<int32, [4]> var_1029 = const()[name = tensor<string, []>("op_1029"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1030_cast_fp16 = reshape(shape = var_1029, x = linear_91_cast_fp16)[name = tensor<string, []>("op_1030_cast_fp16")];
            tensor<int32, [4]> var_1032 = const()[name = tensor<string, []>("op_1032"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1033_cast_fp16 = reshape(shape = var_1032, x = linear_92_cast_fp16)[name = tensor<string, []>("op_1033_cast_fp16")];
            tensor<int32, [4]> value_states_63_perm_0 = const()[name = tensor<string, []>("value_states_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1036_transpose_x_0 = const()[name = tensor<string, []>("op_1036_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1036_transpose_y_0 = const()[name = tensor<string, []>("op_1036_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_111_perm_0 = const()[name = tensor<string, []>("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_112_perm_0 = const()[name = tensor<string, []>("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_112 = transpose(perm = transpose_112_perm_0, x = var_1030_cast_fp16)[name = tensor<string, []>("transpose_180")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_111 = transpose(perm = transpose_111_perm_0, x = var_1027_cast_fp16)[name = tensor<string, []>("transpose_181")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1036_cast_fp16 = matmul(transpose_x = var_1036_transpose_x_0, transpose_y = var_1036_transpose_y_0, x = transpose_111, y = transpose_112)[name = tensor<string, []>("op_1036_cast_fp16")];
            tensor<fp16, []> var_1037_to_fp16 = const()[name = tensor<string, []>("op_1037_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_61_cast_fp16 = mul(x = var_1036_cast_fp16, y = var_1037_to_fp16)[name = tensor<string, []>("attn_weights_61_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1039_cast_fp16 = softmax(axis = var_11, x = attn_weights_61_cast_fp16)[name = tensor<string, []>("op_1039_cast_fp16")];
            tensor<bool, []> attn_output_61_transpose_x_0 = const()[name = tensor<string, []>("attn_output_61_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_61_transpose_y_0 = const()[name = tensor<string, []>("attn_output_61_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_63_cast_fp16 = transpose(perm = value_states_63_perm_0, x = var_1033_cast_fp16)[name = tensor<string, []>("transpose_182")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_61_cast_fp16 = matmul(transpose_x = attn_output_61_transpose_x_0, transpose_y = attn_output_61_transpose_y_0, x = var_1039_cast_fp16, y = value_states_63_cast_fp16)[name = tensor<string, []>("attn_output_61_cast_fp16")];
            tensor<int32, [4]> var_1043_perm_0 = const()[name = tensor<string, []>("op_1043_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1045 = const()[name = tensor<string, []>("op_1045"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1043_cast_fp16 = transpose(perm = var_1043_perm_0, x = attn_output_61_cast_fp16)[name = tensor<string, []>("transpose_179")];
            tensor<fp16, [1, 1024, 1152]> input_213_cast_fp16 = reshape(shape = var_1045, x = var_1043_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_15_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(465175680)))];
            tensor<fp16, [1152]> encoder_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467829952)))];
            tensor<fp16, [1, 1024, 1152]> linear_93_cast_fp16 = linear(bias = encoder_layers_15_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_15_self_attn_out_proj_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_215_cast_fp16 = add(x = input_209_cast_fp16, y = linear_93_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
            tensor<int32, [1]> input_217_axes_0 = const()[name = tensor<string, []>("input_217_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_15_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467832320)))];
            tensor<fp16, [1152]> encoder_layers_15_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467834688)))];
            tensor<fp16, [1, 1024, 1152]> input_217_cast_fp16 = layer_norm(axes = input_217_axes_0, beta = encoder_layers_15_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_layer_norm2_weight_to_fp16, x = input_215_cast_fp16)[name = tensor<string, []>("input_217_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_15_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(467837056)))];
            tensor<fp16, [4304]> encoder_layers_15_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477753536)))];
            tensor<fp16, [1, 1024, 4304]> linear_94_cast_fp16 = linear(bias = encoder_layers_15_mlp_fc1_bias_to_fp16, weight = encoder_layers_15_mlp_fc1_weight_to_fp16, x = input_217_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")];
            tensor<string, []> input_221_mode_0 = const()[name = tensor<string, []>("input_221_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_221_cast_fp16 = gelu(mode = input_221_mode_0, x = linear_94_cast_fp16)[name = tensor<string, []>("input_221_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_15_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477762240)))];
            tensor<fp16, [1152]> encoder_layers_15_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487678720)))];
            tensor<fp16, [1, 1024, 1152]> linear_95_cast_fp16 = linear(bias = encoder_layers_15_mlp_fc2_bias_to_fp16, weight = encoder_layers_15_mlp_fc2_weight_to_fp16, x = input_221_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_223_cast_fp16 = add(x = input_215_cast_fp16, y = linear_95_cast_fp16)[name = tensor<string, []>("input_223_cast_fp16")];
            tensor<int32, [1]> hidden_states_97_axes_0 = const()[name = tensor<string, []>("hidden_states_97_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_16_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487681088)))];
            tensor<fp16, [1152]> encoder_layers_16_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487683456)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_97_cast_fp16 = layer_norm(axes = hidden_states_97_axes_0, beta = encoder_layers_16_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_layer_norm1_weight_to_fp16, x = input_223_cast_fp16)[name = tensor<string, []>("hidden_states_97_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(487685824)))];
            tensor<fp16, [1152]> encoder_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490340096)))];
            tensor<fp16, [1, 1024, 1152]> linear_96_cast_fp16 = linear(bias = encoder_layers_16_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_q_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(490342464)))];
            tensor<fp16, [1152]> encoder_layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492996736)))];
            tensor<fp16, [1, 1024, 1152]> linear_97_cast_fp16 = linear(bias = encoder_layers_16_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_k_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(492999104)))];
            tensor<fp16, [1152]> encoder_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(495653376)))];
            tensor<fp16, [1, 1024, 1152]> linear_98_cast_fp16 = linear(bias = encoder_layers_16_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_v_proj_weight_to_fp16, x = hidden_states_97_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
            tensor<int32, [4]> var_1088 = const()[name = tensor<string, []>("op_1088"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1089_cast_fp16 = reshape(shape = var_1088, x = linear_96_cast_fp16)[name = tensor<string, []>("op_1089_cast_fp16")];
            tensor<int32, [4]> var_1091 = const()[name = tensor<string, []>("op_1091"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1092_cast_fp16 = reshape(shape = var_1091, x = linear_97_cast_fp16)[name = tensor<string, []>("op_1092_cast_fp16")];
            tensor<int32, [4]> var_1094 = const()[name = tensor<string, []>("op_1094"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1095_cast_fp16 = reshape(shape = var_1094, x = linear_98_cast_fp16)[name = tensor<string, []>("op_1095_cast_fp16")];
            tensor<int32, [4]> value_states_67_perm_0 = const()[name = tensor<string, []>("value_states_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1098_transpose_x_0 = const()[name = tensor<string, []>("op_1098_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1098_transpose_y_0 = const()[name = tensor<string, []>("op_1098_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_113_perm_0 = const()[name = tensor<string, []>("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_114_perm_0 = const()[name = tensor<string, []>("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_114 = transpose(perm = transpose_114_perm_0, x = var_1092_cast_fp16)[name = tensor<string, []>("transpose_176")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_113 = transpose(perm = transpose_113_perm_0, x = var_1089_cast_fp16)[name = tensor<string, []>("transpose_177")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1098_cast_fp16 = matmul(transpose_x = var_1098_transpose_x_0, transpose_y = var_1098_transpose_y_0, x = transpose_113, y = transpose_114)[name = tensor<string, []>("op_1098_cast_fp16")];
            tensor<fp16, []> var_1099_to_fp16 = const()[name = tensor<string, []>("op_1099_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_65_cast_fp16 = mul(x = var_1098_cast_fp16, y = var_1099_to_fp16)[name = tensor<string, []>("attn_weights_65_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1101_cast_fp16 = softmax(axis = var_11, x = attn_weights_65_cast_fp16)[name = tensor<string, []>("op_1101_cast_fp16")];
            tensor<bool, []> attn_output_65_transpose_x_0 = const()[name = tensor<string, []>("attn_output_65_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_65_transpose_y_0 = const()[name = tensor<string, []>("attn_output_65_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_67_cast_fp16 = transpose(perm = value_states_67_perm_0, x = var_1095_cast_fp16)[name = tensor<string, []>("transpose_178")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_65_cast_fp16 = matmul(transpose_x = attn_output_65_transpose_x_0, transpose_y = attn_output_65_transpose_y_0, x = var_1101_cast_fp16, y = value_states_67_cast_fp16)[name = tensor<string, []>("attn_output_65_cast_fp16")];
            tensor<int32, [4]> var_1105_perm_0 = const()[name = tensor<string, []>("op_1105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1107 = const()[name = tensor<string, []>("op_1107"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1105_cast_fp16 = transpose(perm = var_1105_perm_0, x = attn_output_65_cast_fp16)[name = tensor<string, []>("transpose_175")];
            tensor<fp16, [1, 1024, 1152]> input_227_cast_fp16 = reshape(shape = var_1107, x = var_1105_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_16_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(495655744)))];
            tensor<fp16, [1152]> encoder_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498310016)))];
            tensor<fp16, [1, 1024, 1152]> linear_99_cast_fp16 = linear(bias = encoder_layers_16_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_16_self_attn_out_proj_weight_to_fp16, x = input_227_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_229_cast_fp16 = add(x = input_223_cast_fp16, y = linear_99_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")];
            tensor<int32, [1]> input_231_axes_0 = const()[name = tensor<string, []>("input_231_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_16_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498312384)))];
            tensor<fp16, [1152]> encoder_layers_16_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498314752)))];
            tensor<fp16, [1, 1024, 1152]> input_231_cast_fp16 = layer_norm(axes = input_231_axes_0, beta = encoder_layers_16_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_layer_norm2_weight_to_fp16, x = input_229_cast_fp16)[name = tensor<string, []>("input_231_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_16_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(498317120)))];
            tensor<fp16, [4304]> encoder_layers_16_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508233600)))];
            tensor<fp16, [1, 1024, 4304]> linear_100_cast_fp16 = linear(bias = encoder_layers_16_mlp_fc1_bias_to_fp16, weight = encoder_layers_16_mlp_fc1_weight_to_fp16, x = input_231_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")];
            tensor<string, []> input_235_mode_0 = const()[name = tensor<string, []>("input_235_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_235_cast_fp16 = gelu(mode = input_235_mode_0, x = linear_100_cast_fp16)[name = tensor<string, []>("input_235_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_16_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(508242304)))];
            tensor<fp16, [1152]> encoder_layers_16_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(518158784)))];
            tensor<fp16, [1, 1024, 1152]> linear_101_cast_fp16 = linear(bias = encoder_layers_16_mlp_fc2_bias_to_fp16, weight = encoder_layers_16_mlp_fc2_weight_to_fp16, x = input_235_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_237_cast_fp16 = add(x = input_229_cast_fp16, y = linear_101_cast_fp16)[name = tensor<string, []>("input_237_cast_fp16")];
            tensor<int32, [1]> hidden_states_103_axes_0 = const()[name = tensor<string, []>("hidden_states_103_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_17_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(518161152)))];
            tensor<fp16, [1152]> encoder_layers_17_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(518163520)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_103_cast_fp16 = layer_norm(axes = hidden_states_103_axes_0, beta = encoder_layers_17_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_17_layer_norm1_weight_to_fp16, x = input_237_cast_fp16)[name = tensor<string, []>("hidden_states_103_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(518165888)))];
            tensor<fp16, [1152]> encoder_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520820160)))];
            tensor<fp16, [1, 1024, 1152]> linear_102_cast_fp16 = linear(bias = encoder_layers_17_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_q_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520822528)))];
            tensor<fp16, [1152]> encoder_layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(523476800)))];
            tensor<fp16, [1, 1024, 1152]> linear_103_cast_fp16 = linear(bias = encoder_layers_17_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_k_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(523479168)))];
            tensor<fp16, [1152]> encoder_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526133440)))];
            tensor<fp16, [1, 1024, 1152]> linear_104_cast_fp16 = linear(bias = encoder_layers_17_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_v_proj_weight_to_fp16, x = hidden_states_103_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
            tensor<int32, [4]> var_1150 = const()[name = tensor<string, []>("op_1150"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1151_cast_fp16 = reshape(shape = var_1150, x = linear_102_cast_fp16)[name = tensor<string, []>("op_1151_cast_fp16")];
            tensor<int32, [4]> var_1153 = const()[name = tensor<string, []>("op_1153"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1154_cast_fp16 = reshape(shape = var_1153, x = linear_103_cast_fp16)[name = tensor<string, []>("op_1154_cast_fp16")];
            tensor<int32, [4]> var_1156 = const()[name = tensor<string, []>("op_1156"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1157_cast_fp16 = reshape(shape = var_1156, x = linear_104_cast_fp16)[name = tensor<string, []>("op_1157_cast_fp16")];
            tensor<int32, [4]> value_states_71_perm_0 = const()[name = tensor<string, []>("value_states_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1160_transpose_x_0 = const()[name = tensor<string, []>("op_1160_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1160_transpose_y_0 = const()[name = tensor<string, []>("op_1160_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_115_perm_0 = const()[name = tensor<string, []>("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_116_perm_0 = const()[name = tensor<string, []>("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_116 = transpose(perm = transpose_116_perm_0, x = var_1154_cast_fp16)[name = tensor<string, []>("transpose_172")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_115 = transpose(perm = transpose_115_perm_0, x = var_1151_cast_fp16)[name = tensor<string, []>("transpose_173")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1160_cast_fp16 = matmul(transpose_x = var_1160_transpose_x_0, transpose_y = var_1160_transpose_y_0, x = transpose_115, y = transpose_116)[name = tensor<string, []>("op_1160_cast_fp16")];
            tensor<fp16, []> var_1161_to_fp16 = const()[name = tensor<string, []>("op_1161_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_69_cast_fp16 = mul(x = var_1160_cast_fp16, y = var_1161_to_fp16)[name = tensor<string, []>("attn_weights_69_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1163_cast_fp16 = softmax(axis = var_11, x = attn_weights_69_cast_fp16)[name = tensor<string, []>("op_1163_cast_fp16")];
            tensor<bool, []> attn_output_69_transpose_x_0 = const()[name = tensor<string, []>("attn_output_69_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_69_transpose_y_0 = const()[name = tensor<string, []>("attn_output_69_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_71_cast_fp16 = transpose(perm = value_states_71_perm_0, x = var_1157_cast_fp16)[name = tensor<string, []>("transpose_174")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_69_cast_fp16 = matmul(transpose_x = attn_output_69_transpose_x_0, transpose_y = attn_output_69_transpose_y_0, x = var_1163_cast_fp16, y = value_states_71_cast_fp16)[name = tensor<string, []>("attn_output_69_cast_fp16")];
            tensor<int32, [4]> var_1167_perm_0 = const()[name = tensor<string, []>("op_1167_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1169 = const()[name = tensor<string, []>("op_1169"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1167_cast_fp16 = transpose(perm = var_1167_perm_0, x = attn_output_69_cast_fp16)[name = tensor<string, []>("transpose_171")];
            tensor<fp16, [1, 1024, 1152]> input_241_cast_fp16 = reshape(shape = var_1169, x = var_1167_cast_fp16)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_17_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526135808)))];
            tensor<fp16, [1152]> encoder_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528790080)))];
            tensor<fp16, [1, 1024, 1152]> linear_105_cast_fp16 = linear(bias = encoder_layers_17_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_17_self_attn_out_proj_weight_to_fp16, x = input_241_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_243_cast_fp16 = add(x = input_237_cast_fp16, y = linear_105_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")];
            tensor<int32, [1]> input_245_axes_0 = const()[name = tensor<string, []>("input_245_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_17_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528792448)))];
            tensor<fp16, [1152]> encoder_layers_17_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528794816)))];
            tensor<fp16, [1, 1024, 1152]> input_245_cast_fp16 = layer_norm(axes = input_245_axes_0, beta = encoder_layers_17_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_17_layer_norm2_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("input_245_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_17_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528797184)))];
            tensor<fp16, [4304]> encoder_layers_17_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538713664)))];
            tensor<fp16, [1, 1024, 4304]> linear_106_cast_fp16 = linear(bias = encoder_layers_17_mlp_fc1_bias_to_fp16, weight = encoder_layers_17_mlp_fc1_weight_to_fp16, x = input_245_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")];
            tensor<string, []> input_249_mode_0 = const()[name = tensor<string, []>("input_249_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_249_cast_fp16 = gelu(mode = input_249_mode_0, x = linear_106_cast_fp16)[name = tensor<string, []>("input_249_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_17_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(538722368)))];
            tensor<fp16, [1152]> encoder_layers_17_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_17_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(548638848)))];
            tensor<fp16, [1, 1024, 1152]> linear_107_cast_fp16 = linear(bias = encoder_layers_17_mlp_fc2_bias_to_fp16, weight = encoder_layers_17_mlp_fc2_weight_to_fp16, x = input_249_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_251_cast_fp16 = add(x = input_243_cast_fp16, y = linear_107_cast_fp16)[name = tensor<string, []>("input_251_cast_fp16")];
            tensor<int32, [1]> hidden_states_109_axes_0 = const()[name = tensor<string, []>("hidden_states_109_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_18_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(548641216)))];
            tensor<fp16, [1152]> encoder_layers_18_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(548643584)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_109_cast_fp16 = layer_norm(axes = hidden_states_109_axes_0, beta = encoder_layers_18_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_18_layer_norm1_weight_to_fp16, x = input_251_cast_fp16)[name = tensor<string, []>("hidden_states_109_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(548645952)))];
            tensor<fp16, [1152]> encoder_layers_18_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(551300224)))];
            tensor<fp16, [1, 1024, 1152]> linear_108_cast_fp16 = linear(bias = encoder_layers_18_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_q_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(551302592)))];
            tensor<fp16, [1152]> encoder_layers_18_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(553956864)))];
            tensor<fp16, [1, 1024, 1152]> linear_109_cast_fp16 = linear(bias = encoder_layers_18_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_k_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(553959232)))];
            tensor<fp16, [1152]> encoder_layers_18_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(556613504)))];
            tensor<fp16, [1, 1024, 1152]> linear_110_cast_fp16 = linear(bias = encoder_layers_18_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_v_proj_weight_to_fp16, x = hidden_states_109_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
            tensor<int32, [4]> var_1212 = const()[name = tensor<string, []>("op_1212"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1213_cast_fp16 = reshape(shape = var_1212, x = linear_108_cast_fp16)[name = tensor<string, []>("op_1213_cast_fp16")];
            tensor<int32, [4]> var_1215 = const()[name = tensor<string, []>("op_1215"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1216_cast_fp16 = reshape(shape = var_1215, x = linear_109_cast_fp16)[name = tensor<string, []>("op_1216_cast_fp16")];
            tensor<int32, [4]> var_1218 = const()[name = tensor<string, []>("op_1218"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1219_cast_fp16 = reshape(shape = var_1218, x = linear_110_cast_fp16)[name = tensor<string, []>("op_1219_cast_fp16")];
            tensor<int32, [4]> value_states_75_perm_0 = const()[name = tensor<string, []>("value_states_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1222_transpose_x_0 = const()[name = tensor<string, []>("op_1222_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1222_transpose_y_0 = const()[name = tensor<string, []>("op_1222_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_117_perm_0 = const()[name = tensor<string, []>("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_118_perm_0 = const()[name = tensor<string, []>("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_118 = transpose(perm = transpose_118_perm_0, x = var_1216_cast_fp16)[name = tensor<string, []>("transpose_168")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_117 = transpose(perm = transpose_117_perm_0, x = var_1213_cast_fp16)[name = tensor<string, []>("transpose_169")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1222_cast_fp16 = matmul(transpose_x = var_1222_transpose_x_0, transpose_y = var_1222_transpose_y_0, x = transpose_117, y = transpose_118)[name = tensor<string, []>("op_1222_cast_fp16")];
            tensor<fp16, []> var_1223_to_fp16 = const()[name = tensor<string, []>("op_1223_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_73_cast_fp16 = mul(x = var_1222_cast_fp16, y = var_1223_to_fp16)[name = tensor<string, []>("attn_weights_73_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1225_cast_fp16 = softmax(axis = var_11, x = attn_weights_73_cast_fp16)[name = tensor<string, []>("op_1225_cast_fp16")];
            tensor<bool, []> attn_output_73_transpose_x_0 = const()[name = tensor<string, []>("attn_output_73_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_73_transpose_y_0 = const()[name = tensor<string, []>("attn_output_73_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_75_cast_fp16 = transpose(perm = value_states_75_perm_0, x = var_1219_cast_fp16)[name = tensor<string, []>("transpose_170")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_73_cast_fp16 = matmul(transpose_x = attn_output_73_transpose_x_0, transpose_y = attn_output_73_transpose_y_0, x = var_1225_cast_fp16, y = value_states_75_cast_fp16)[name = tensor<string, []>("attn_output_73_cast_fp16")];
            tensor<int32, [4]> var_1229_perm_0 = const()[name = tensor<string, []>("op_1229_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1231 = const()[name = tensor<string, []>("op_1231"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1229_cast_fp16 = transpose(perm = var_1229_perm_0, x = attn_output_73_cast_fp16)[name = tensor<string, []>("transpose_167")];
            tensor<fp16, [1, 1024, 1152]> input_255_cast_fp16 = reshape(shape = var_1231, x = var_1229_cast_fp16)[name = tensor<string, []>("input_255_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_18_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(556615872)))];
            tensor<fp16, [1152]> encoder_layers_18_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559270144)))];
            tensor<fp16, [1, 1024, 1152]> linear_111_cast_fp16 = linear(bias = encoder_layers_18_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_18_self_attn_out_proj_weight_to_fp16, x = input_255_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_257_cast_fp16 = add(x = input_251_cast_fp16, y = linear_111_cast_fp16)[name = tensor<string, []>("input_257_cast_fp16")];
            tensor<int32, [1]> input_259_axes_0 = const()[name = tensor<string, []>("input_259_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_18_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559272512)))];
            tensor<fp16, [1152]> encoder_layers_18_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559274880)))];
            tensor<fp16, [1, 1024, 1152]> input_259_cast_fp16 = layer_norm(axes = input_259_axes_0, beta = encoder_layers_18_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_18_layer_norm2_weight_to_fp16, x = input_257_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_18_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(559277248)))];
            tensor<fp16, [4304]> encoder_layers_18_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569193728)))];
            tensor<fp16, [1, 1024, 4304]> linear_112_cast_fp16 = linear(bias = encoder_layers_18_mlp_fc1_bias_to_fp16, weight = encoder_layers_18_mlp_fc1_weight_to_fp16, x = input_259_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")];
            tensor<string, []> input_263_mode_0 = const()[name = tensor<string, []>("input_263_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_263_cast_fp16 = gelu(mode = input_263_mode_0, x = linear_112_cast_fp16)[name = tensor<string, []>("input_263_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_18_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569202432)))];
            tensor<fp16, [1152]> encoder_layers_18_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_18_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579118912)))];
            tensor<fp16, [1, 1024, 1152]> linear_113_cast_fp16 = linear(bias = encoder_layers_18_mlp_fc2_bias_to_fp16, weight = encoder_layers_18_mlp_fc2_weight_to_fp16, x = input_263_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_265_cast_fp16 = add(x = input_257_cast_fp16, y = linear_113_cast_fp16)[name = tensor<string, []>("input_265_cast_fp16")];
            tensor<int32, [1]> hidden_states_115_axes_0 = const()[name = tensor<string, []>("hidden_states_115_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_19_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579121280)))];
            tensor<fp16, [1152]> encoder_layers_19_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579123648)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_115_cast_fp16 = layer_norm(axes = hidden_states_115_axes_0, beta = encoder_layers_19_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_19_layer_norm1_weight_to_fp16, x = input_265_cast_fp16)[name = tensor<string, []>("hidden_states_115_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(579126016)))];
            tensor<fp16, [1152]> encoder_layers_19_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(581780288)))];
            tensor<fp16, [1, 1024, 1152]> linear_114_cast_fp16 = linear(bias = encoder_layers_19_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_q_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(581782656)))];
            tensor<fp16, [1152]> encoder_layers_19_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(584436928)))];
            tensor<fp16, [1, 1024, 1152]> linear_115_cast_fp16 = linear(bias = encoder_layers_19_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_k_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(584439296)))];
            tensor<fp16, [1152]> encoder_layers_19_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(587093568)))];
            tensor<fp16, [1, 1024, 1152]> linear_116_cast_fp16 = linear(bias = encoder_layers_19_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_v_proj_weight_to_fp16, x = hidden_states_115_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
            tensor<int32, [4]> var_1274 = const()[name = tensor<string, []>("op_1274"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1275_cast_fp16 = reshape(shape = var_1274, x = linear_114_cast_fp16)[name = tensor<string, []>("op_1275_cast_fp16")];
            tensor<int32, [4]> var_1277 = const()[name = tensor<string, []>("op_1277"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1278_cast_fp16 = reshape(shape = var_1277, x = linear_115_cast_fp16)[name = tensor<string, []>("op_1278_cast_fp16")];
            tensor<int32, [4]> var_1280 = const()[name = tensor<string, []>("op_1280"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1281_cast_fp16 = reshape(shape = var_1280, x = linear_116_cast_fp16)[name = tensor<string, []>("op_1281_cast_fp16")];
            tensor<int32, [4]> value_states_79_perm_0 = const()[name = tensor<string, []>("value_states_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1284_transpose_x_0 = const()[name = tensor<string, []>("op_1284_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1284_transpose_y_0 = const()[name = tensor<string, []>("op_1284_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_119_perm_0 = const()[name = tensor<string, []>("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_120_perm_0 = const()[name = tensor<string, []>("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_120 = transpose(perm = transpose_120_perm_0, x = var_1278_cast_fp16)[name = tensor<string, []>("transpose_164")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_119 = transpose(perm = transpose_119_perm_0, x = var_1275_cast_fp16)[name = tensor<string, []>("transpose_165")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1284_cast_fp16 = matmul(transpose_x = var_1284_transpose_x_0, transpose_y = var_1284_transpose_y_0, x = transpose_119, y = transpose_120)[name = tensor<string, []>("op_1284_cast_fp16")];
            tensor<fp16, []> var_1285_to_fp16 = const()[name = tensor<string, []>("op_1285_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_77_cast_fp16 = mul(x = var_1284_cast_fp16, y = var_1285_to_fp16)[name = tensor<string, []>("attn_weights_77_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1287_cast_fp16 = softmax(axis = var_11, x = attn_weights_77_cast_fp16)[name = tensor<string, []>("op_1287_cast_fp16")];
            tensor<bool, []> attn_output_77_transpose_x_0 = const()[name = tensor<string, []>("attn_output_77_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_77_transpose_y_0 = const()[name = tensor<string, []>("attn_output_77_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_79_cast_fp16 = transpose(perm = value_states_79_perm_0, x = var_1281_cast_fp16)[name = tensor<string, []>("transpose_166")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_77_cast_fp16 = matmul(transpose_x = attn_output_77_transpose_x_0, transpose_y = attn_output_77_transpose_y_0, x = var_1287_cast_fp16, y = value_states_79_cast_fp16)[name = tensor<string, []>("attn_output_77_cast_fp16")];
            tensor<int32, [4]> var_1291_perm_0 = const()[name = tensor<string, []>("op_1291_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1293 = const()[name = tensor<string, []>("op_1293"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1291_cast_fp16 = transpose(perm = var_1291_perm_0, x = attn_output_77_cast_fp16)[name = tensor<string, []>("transpose_163")];
            tensor<fp16, [1, 1024, 1152]> input_269_cast_fp16 = reshape(shape = var_1293, x = var_1291_cast_fp16)[name = tensor<string, []>("input_269_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_19_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(587095936)))];
            tensor<fp16, [1152]> encoder_layers_19_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589750208)))];
            tensor<fp16, [1, 1024, 1152]> linear_117_cast_fp16 = linear(bias = encoder_layers_19_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_19_self_attn_out_proj_weight_to_fp16, x = input_269_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_271_cast_fp16 = add(x = input_265_cast_fp16, y = linear_117_cast_fp16)[name = tensor<string, []>("input_271_cast_fp16")];
            tensor<int32, [1]> input_273_axes_0 = const()[name = tensor<string, []>("input_273_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_19_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589752576)))];
            tensor<fp16, [1152]> encoder_layers_19_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589754944)))];
            tensor<fp16, [1, 1024, 1152]> input_273_cast_fp16 = layer_norm(axes = input_273_axes_0, beta = encoder_layers_19_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_19_layer_norm2_weight_to_fp16, x = input_271_cast_fp16)[name = tensor<string, []>("input_273_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_19_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(589757312)))];
            tensor<fp16, [4304]> encoder_layers_19_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(599673792)))];
            tensor<fp16, [1, 1024, 4304]> linear_118_cast_fp16 = linear(bias = encoder_layers_19_mlp_fc1_bias_to_fp16, weight = encoder_layers_19_mlp_fc1_weight_to_fp16, x = input_273_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")];
            tensor<string, []> input_277_mode_0 = const()[name = tensor<string, []>("input_277_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_277_cast_fp16 = gelu(mode = input_277_mode_0, x = linear_118_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_19_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(599682496)))];
            tensor<fp16, [1152]> encoder_layers_19_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_19_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609598976)))];
            tensor<fp16, [1, 1024, 1152]> linear_119_cast_fp16 = linear(bias = encoder_layers_19_mlp_fc2_bias_to_fp16, weight = encoder_layers_19_mlp_fc2_weight_to_fp16, x = input_277_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_279_cast_fp16 = add(x = input_271_cast_fp16, y = linear_119_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")];
            tensor<int32, [1]> hidden_states_121_axes_0 = const()[name = tensor<string, []>("hidden_states_121_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_20_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609601344)))];
            tensor<fp16, [1152]> encoder_layers_20_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609603712)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_121_cast_fp16 = layer_norm(axes = hidden_states_121_axes_0, beta = encoder_layers_20_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_20_layer_norm1_weight_to_fp16, x = input_279_cast_fp16)[name = tensor<string, []>("hidden_states_121_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(609606080)))];
            tensor<fp16, [1152]> encoder_layers_20_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612260352)))];
            tensor<fp16, [1, 1024, 1152]> linear_120_cast_fp16 = linear(bias = encoder_layers_20_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_q_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(612262720)))];
            tensor<fp16, [1152]> encoder_layers_20_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614916992)))];
            tensor<fp16, [1, 1024, 1152]> linear_121_cast_fp16 = linear(bias = encoder_layers_20_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_k_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614919360)))];
            tensor<fp16, [1152]> encoder_layers_20_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617573632)))];
            tensor<fp16, [1, 1024, 1152]> linear_122_cast_fp16 = linear(bias = encoder_layers_20_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_v_proj_weight_to_fp16, x = hidden_states_121_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
            tensor<int32, [4]> var_1336 = const()[name = tensor<string, []>("op_1336"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1337_cast_fp16 = reshape(shape = var_1336, x = linear_120_cast_fp16)[name = tensor<string, []>("op_1337_cast_fp16")];
            tensor<int32, [4]> var_1339 = const()[name = tensor<string, []>("op_1339"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1340_cast_fp16 = reshape(shape = var_1339, x = linear_121_cast_fp16)[name = tensor<string, []>("op_1340_cast_fp16")];
            tensor<int32, [4]> var_1342 = const()[name = tensor<string, []>("op_1342"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1343_cast_fp16 = reshape(shape = var_1342, x = linear_122_cast_fp16)[name = tensor<string, []>("op_1343_cast_fp16")];
            tensor<int32, [4]> value_states_83_perm_0 = const()[name = tensor<string, []>("value_states_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1346_transpose_x_0 = const()[name = tensor<string, []>("op_1346_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1346_transpose_y_0 = const()[name = tensor<string, []>("op_1346_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_121_perm_0 = const()[name = tensor<string, []>("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_122_perm_0 = const()[name = tensor<string, []>("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_122 = transpose(perm = transpose_122_perm_0, x = var_1340_cast_fp16)[name = tensor<string, []>("transpose_160")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_121 = transpose(perm = transpose_121_perm_0, x = var_1337_cast_fp16)[name = tensor<string, []>("transpose_161")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1346_cast_fp16 = matmul(transpose_x = var_1346_transpose_x_0, transpose_y = var_1346_transpose_y_0, x = transpose_121, y = transpose_122)[name = tensor<string, []>("op_1346_cast_fp16")];
            tensor<fp16, []> var_1347_to_fp16 = const()[name = tensor<string, []>("op_1347_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_81_cast_fp16 = mul(x = var_1346_cast_fp16, y = var_1347_to_fp16)[name = tensor<string, []>("attn_weights_81_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1349_cast_fp16 = softmax(axis = var_11, x = attn_weights_81_cast_fp16)[name = tensor<string, []>("op_1349_cast_fp16")];
            tensor<bool, []> attn_output_81_transpose_x_0 = const()[name = tensor<string, []>("attn_output_81_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_81_transpose_y_0 = const()[name = tensor<string, []>("attn_output_81_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_83_cast_fp16 = transpose(perm = value_states_83_perm_0, x = var_1343_cast_fp16)[name = tensor<string, []>("transpose_162")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_81_cast_fp16 = matmul(transpose_x = attn_output_81_transpose_x_0, transpose_y = attn_output_81_transpose_y_0, x = var_1349_cast_fp16, y = value_states_83_cast_fp16)[name = tensor<string, []>("attn_output_81_cast_fp16")];
            tensor<int32, [4]> var_1353_perm_0 = const()[name = tensor<string, []>("op_1353_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1355 = const()[name = tensor<string, []>("op_1355"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1353_cast_fp16 = transpose(perm = var_1353_perm_0, x = attn_output_81_cast_fp16)[name = tensor<string, []>("transpose_159")];
            tensor<fp16, [1, 1024, 1152]> input_283_cast_fp16 = reshape(shape = var_1355, x = var_1353_cast_fp16)[name = tensor<string, []>("input_283_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_20_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617576000)))];
            tensor<fp16, [1152]> encoder_layers_20_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(620230272)))];
            tensor<fp16, [1, 1024, 1152]> linear_123_cast_fp16 = linear(bias = encoder_layers_20_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_20_self_attn_out_proj_weight_to_fp16, x = input_283_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_285_cast_fp16 = add(x = input_279_cast_fp16, y = linear_123_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")];
            tensor<int32, [1]> input_287_axes_0 = const()[name = tensor<string, []>("input_287_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_20_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(620232640)))];
            tensor<fp16, [1152]> encoder_layers_20_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(620235008)))];
            tensor<fp16, [1, 1024, 1152]> input_287_cast_fp16 = layer_norm(axes = input_287_axes_0, beta = encoder_layers_20_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_20_layer_norm2_weight_to_fp16, x = input_285_cast_fp16)[name = tensor<string, []>("input_287_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_20_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(620237376)))];
            tensor<fp16, [4304]> encoder_layers_20_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630153856)))];
            tensor<fp16, [1, 1024, 4304]> linear_124_cast_fp16 = linear(bias = encoder_layers_20_mlp_fc1_bias_to_fp16, weight = encoder_layers_20_mlp_fc1_weight_to_fp16, x = input_287_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")];
            tensor<string, []> input_291_mode_0 = const()[name = tensor<string, []>("input_291_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_291_cast_fp16 = gelu(mode = input_291_mode_0, x = linear_124_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_20_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630162560)))];
            tensor<fp16, [1152]> encoder_layers_20_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_20_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640079040)))];
            tensor<fp16, [1, 1024, 1152]> linear_125_cast_fp16 = linear(bias = encoder_layers_20_mlp_fc2_bias_to_fp16, weight = encoder_layers_20_mlp_fc2_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_293_cast_fp16 = add(x = input_285_cast_fp16, y = linear_125_cast_fp16)[name = tensor<string, []>("input_293_cast_fp16")];
            tensor<int32, [1]> hidden_states_127_axes_0 = const()[name = tensor<string, []>("hidden_states_127_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_21_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640081408)))];
            tensor<fp16, [1152]> encoder_layers_21_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640083776)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_127_cast_fp16 = layer_norm(axes = hidden_states_127_axes_0, beta = encoder_layers_21_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_21_layer_norm1_weight_to_fp16, x = input_293_cast_fp16)[name = tensor<string, []>("hidden_states_127_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(640086144)))];
            tensor<fp16, [1152]> encoder_layers_21_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642740416)))];
            tensor<fp16, [1, 1024, 1152]> linear_126_cast_fp16 = linear(bias = encoder_layers_21_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_q_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(642742784)))];
            tensor<fp16, [1152]> encoder_layers_21_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645397056)))];
            tensor<fp16, [1, 1024, 1152]> linear_127_cast_fp16 = linear(bias = encoder_layers_21_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_k_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(645399424)))];
            tensor<fp16, [1152]> encoder_layers_21_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(648053696)))];
            tensor<fp16, [1, 1024, 1152]> linear_128_cast_fp16 = linear(bias = encoder_layers_21_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_v_proj_weight_to_fp16, x = hidden_states_127_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")];
            tensor<int32, [4]> var_1398 = const()[name = tensor<string, []>("op_1398"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1399_cast_fp16 = reshape(shape = var_1398, x = linear_126_cast_fp16)[name = tensor<string, []>("op_1399_cast_fp16")];
            tensor<int32, [4]> var_1401 = const()[name = tensor<string, []>("op_1401"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1402_cast_fp16 = reshape(shape = var_1401, x = linear_127_cast_fp16)[name = tensor<string, []>("op_1402_cast_fp16")];
            tensor<int32, [4]> var_1404 = const()[name = tensor<string, []>("op_1404"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1405_cast_fp16 = reshape(shape = var_1404, x = linear_128_cast_fp16)[name = tensor<string, []>("op_1405_cast_fp16")];
            tensor<int32, [4]> value_states_87_perm_0 = const()[name = tensor<string, []>("value_states_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1408_transpose_x_0 = const()[name = tensor<string, []>("op_1408_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1408_transpose_y_0 = const()[name = tensor<string, []>("op_1408_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_123_perm_0 = const()[name = tensor<string, []>("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_124_perm_0 = const()[name = tensor<string, []>("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_124 = transpose(perm = transpose_124_perm_0, x = var_1402_cast_fp16)[name = tensor<string, []>("transpose_156")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_123 = transpose(perm = transpose_123_perm_0, x = var_1399_cast_fp16)[name = tensor<string, []>("transpose_157")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1408_cast_fp16 = matmul(transpose_x = var_1408_transpose_x_0, transpose_y = var_1408_transpose_y_0, x = transpose_123, y = transpose_124)[name = tensor<string, []>("op_1408_cast_fp16")];
            tensor<fp16, []> var_1409_to_fp16 = const()[name = tensor<string, []>("op_1409_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_85_cast_fp16 = mul(x = var_1408_cast_fp16, y = var_1409_to_fp16)[name = tensor<string, []>("attn_weights_85_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1411_cast_fp16 = softmax(axis = var_11, x = attn_weights_85_cast_fp16)[name = tensor<string, []>("op_1411_cast_fp16")];
            tensor<bool, []> attn_output_85_transpose_x_0 = const()[name = tensor<string, []>("attn_output_85_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_85_transpose_y_0 = const()[name = tensor<string, []>("attn_output_85_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_87_cast_fp16 = transpose(perm = value_states_87_perm_0, x = var_1405_cast_fp16)[name = tensor<string, []>("transpose_158")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_85_cast_fp16 = matmul(transpose_x = attn_output_85_transpose_x_0, transpose_y = attn_output_85_transpose_y_0, x = var_1411_cast_fp16, y = value_states_87_cast_fp16)[name = tensor<string, []>("attn_output_85_cast_fp16")];
            tensor<int32, [4]> var_1415_perm_0 = const()[name = tensor<string, []>("op_1415_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1417 = const()[name = tensor<string, []>("op_1417"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1415_cast_fp16 = transpose(perm = var_1415_perm_0, x = attn_output_85_cast_fp16)[name = tensor<string, []>("transpose_155")];
            tensor<fp16, [1, 1024, 1152]> input_297_cast_fp16 = reshape(shape = var_1417, x = var_1415_cast_fp16)[name = tensor<string, []>("input_297_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_21_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(648056064)))];
            tensor<fp16, [1152]> encoder_layers_21_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650710336)))];
            tensor<fp16, [1, 1024, 1152]> linear_129_cast_fp16 = linear(bias = encoder_layers_21_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_21_self_attn_out_proj_weight_to_fp16, x = input_297_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_299_cast_fp16 = add(x = input_293_cast_fp16, y = linear_129_cast_fp16)[name = tensor<string, []>("input_299_cast_fp16")];
            tensor<int32, [1]> input_301_axes_0 = const()[name = tensor<string, []>("input_301_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_21_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650712704)))];
            tensor<fp16, [1152]> encoder_layers_21_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650715072)))];
            tensor<fp16, [1, 1024, 1152]> input_301_cast_fp16 = layer_norm(axes = input_301_axes_0, beta = encoder_layers_21_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_21_layer_norm2_weight_to_fp16, x = input_299_cast_fp16)[name = tensor<string, []>("input_301_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_21_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650717440)))];
            tensor<fp16, [4304]> encoder_layers_21_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(660633920)))];
            tensor<fp16, [1, 1024, 4304]> linear_130_cast_fp16 = linear(bias = encoder_layers_21_mlp_fc1_bias_to_fp16, weight = encoder_layers_21_mlp_fc1_weight_to_fp16, x = input_301_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")];
            tensor<string, []> input_305_mode_0 = const()[name = tensor<string, []>("input_305_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_305_cast_fp16 = gelu(mode = input_305_mode_0, x = linear_130_cast_fp16)[name = tensor<string, []>("input_305_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_21_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(660642624)))];
            tensor<fp16, [1152]> encoder_layers_21_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_21_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670559104)))];
            tensor<fp16, [1, 1024, 1152]> linear_131_cast_fp16 = linear(bias = encoder_layers_21_mlp_fc2_bias_to_fp16, weight = encoder_layers_21_mlp_fc2_weight_to_fp16, x = input_305_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_307_cast_fp16 = add(x = input_299_cast_fp16, y = linear_131_cast_fp16)[name = tensor<string, []>("input_307_cast_fp16")];
            tensor<int32, [1]> hidden_states_133_axes_0 = const()[name = tensor<string, []>("hidden_states_133_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_22_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670561472)))];
            tensor<fp16, [1152]> encoder_layers_22_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670563840)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_133_cast_fp16 = layer_norm(axes = hidden_states_133_axes_0, beta = encoder_layers_22_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_22_layer_norm1_weight_to_fp16, x = input_307_cast_fp16)[name = tensor<string, []>("hidden_states_133_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670566208)))];
            tensor<fp16, [1152]> encoder_layers_22_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673220480)))];
            tensor<fp16, [1, 1024, 1152]> linear_132_cast_fp16 = linear(bias = encoder_layers_22_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_q_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_132_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(673222848)))];
            tensor<fp16, [1152]> encoder_layers_22_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(675877120)))];
            tensor<fp16, [1, 1024, 1152]> linear_133_cast_fp16 = linear(bias = encoder_layers_22_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_k_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(675879488)))];
            tensor<fp16, [1152]> encoder_layers_22_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(678533760)))];
            tensor<fp16, [1, 1024, 1152]> linear_134_cast_fp16 = linear(bias = encoder_layers_22_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_v_proj_weight_to_fp16, x = hidden_states_133_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")];
            tensor<int32, [4]> var_1460 = const()[name = tensor<string, []>("op_1460"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1461_cast_fp16 = reshape(shape = var_1460, x = linear_132_cast_fp16)[name = tensor<string, []>("op_1461_cast_fp16")];
            tensor<int32, [4]> var_1463 = const()[name = tensor<string, []>("op_1463"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1464_cast_fp16 = reshape(shape = var_1463, x = linear_133_cast_fp16)[name = tensor<string, []>("op_1464_cast_fp16")];
            tensor<int32, [4]> var_1466 = const()[name = tensor<string, []>("op_1466"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1467_cast_fp16 = reshape(shape = var_1466, x = linear_134_cast_fp16)[name = tensor<string, []>("op_1467_cast_fp16")];
            tensor<int32, [4]> value_states_91_perm_0 = const()[name = tensor<string, []>("value_states_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1470_transpose_x_0 = const()[name = tensor<string, []>("op_1470_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1470_transpose_y_0 = const()[name = tensor<string, []>("op_1470_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_125_perm_0 = const()[name = tensor<string, []>("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_126_perm_0 = const()[name = tensor<string, []>("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_126 = transpose(perm = transpose_126_perm_0, x = var_1464_cast_fp16)[name = tensor<string, []>("transpose_152")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_125 = transpose(perm = transpose_125_perm_0, x = var_1461_cast_fp16)[name = tensor<string, []>("transpose_153")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1470_cast_fp16 = matmul(transpose_x = var_1470_transpose_x_0, transpose_y = var_1470_transpose_y_0, x = transpose_125, y = transpose_126)[name = tensor<string, []>("op_1470_cast_fp16")];
            tensor<fp16, []> var_1471_to_fp16 = const()[name = tensor<string, []>("op_1471_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_89_cast_fp16 = mul(x = var_1470_cast_fp16, y = var_1471_to_fp16)[name = tensor<string, []>("attn_weights_89_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1473_cast_fp16 = softmax(axis = var_11, x = attn_weights_89_cast_fp16)[name = tensor<string, []>("op_1473_cast_fp16")];
            tensor<bool, []> attn_output_89_transpose_x_0 = const()[name = tensor<string, []>("attn_output_89_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_89_transpose_y_0 = const()[name = tensor<string, []>("attn_output_89_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_91_cast_fp16 = transpose(perm = value_states_91_perm_0, x = var_1467_cast_fp16)[name = tensor<string, []>("transpose_154")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_89_cast_fp16 = matmul(transpose_x = attn_output_89_transpose_x_0, transpose_y = attn_output_89_transpose_y_0, x = var_1473_cast_fp16, y = value_states_91_cast_fp16)[name = tensor<string, []>("attn_output_89_cast_fp16")];
            tensor<int32, [4]> var_1477_perm_0 = const()[name = tensor<string, []>("op_1477_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1479 = const()[name = tensor<string, []>("op_1479"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1477_cast_fp16 = transpose(perm = var_1477_perm_0, x = attn_output_89_cast_fp16)[name = tensor<string, []>("transpose_151")];
            tensor<fp16, [1, 1024, 1152]> input_311_cast_fp16 = reshape(shape = var_1479, x = var_1477_cast_fp16)[name = tensor<string, []>("input_311_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_22_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(678536128)))];
            tensor<fp16, [1152]> encoder_layers_22_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681190400)))];
            tensor<fp16, [1, 1024, 1152]> linear_135_cast_fp16 = linear(bias = encoder_layers_22_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_22_self_attn_out_proj_weight_to_fp16, x = input_311_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_313_cast_fp16 = add(x = input_307_cast_fp16, y = linear_135_cast_fp16)[name = tensor<string, []>("input_313_cast_fp16")];
            tensor<int32, [1]> input_315_axes_0 = const()[name = tensor<string, []>("input_315_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_22_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681192768)))];
            tensor<fp16, [1152]> encoder_layers_22_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681195136)))];
            tensor<fp16, [1, 1024, 1152]> input_315_cast_fp16 = layer_norm(axes = input_315_axes_0, beta = encoder_layers_22_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_22_layer_norm2_weight_to_fp16, x = input_313_cast_fp16)[name = tensor<string, []>("input_315_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_22_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(681197504)))];
            tensor<fp16, [4304]> encoder_layers_22_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691113984)))];
            tensor<fp16, [1, 1024, 4304]> linear_136_cast_fp16 = linear(bias = encoder_layers_22_mlp_fc1_bias_to_fp16, weight = encoder_layers_22_mlp_fc1_weight_to_fp16, x = input_315_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")];
            tensor<string, []> input_319_mode_0 = const()[name = tensor<string, []>("input_319_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_319_cast_fp16 = gelu(mode = input_319_mode_0, x = linear_136_cast_fp16)[name = tensor<string, []>("input_319_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_22_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691122688)))];
            tensor<fp16, [1152]> encoder_layers_22_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_22_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(701039168)))];
            tensor<fp16, [1, 1024, 1152]> linear_137_cast_fp16 = linear(bias = encoder_layers_22_mlp_fc2_bias_to_fp16, weight = encoder_layers_22_mlp_fc2_weight_to_fp16, x = input_319_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_321_cast_fp16 = add(x = input_313_cast_fp16, y = linear_137_cast_fp16)[name = tensor<string, []>("input_321_cast_fp16")];
            tensor<int32, [1]> hidden_states_139_axes_0 = const()[name = tensor<string, []>("hidden_states_139_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_23_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(701041536)))];
            tensor<fp16, [1152]> encoder_layers_23_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(701043904)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_139_cast_fp16 = layer_norm(axes = hidden_states_139_axes_0, beta = encoder_layers_23_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_23_layer_norm1_weight_to_fp16, x = input_321_cast_fp16)[name = tensor<string, []>("hidden_states_139_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(701046272)))];
            tensor<fp16, [1152]> encoder_layers_23_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703700544)))];
            tensor<fp16, [1, 1024, 1152]> linear_138_cast_fp16 = linear(bias = encoder_layers_23_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_q_proj_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(703702912)))];
            tensor<fp16, [1152]> encoder_layers_23_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(706357184)))];
            tensor<fp16, [1, 1024, 1152]> linear_139_cast_fp16 = linear(bias = encoder_layers_23_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_k_proj_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(706359552)))];
            tensor<fp16, [1152]> encoder_layers_23_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709013824)))];
            tensor<fp16, [1, 1024, 1152]> linear_140_cast_fp16 = linear(bias = encoder_layers_23_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_v_proj_weight_to_fp16, x = hidden_states_139_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")];
            tensor<int32, [4]> var_1522 = const()[name = tensor<string, []>("op_1522"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1523_cast_fp16 = reshape(shape = var_1522, x = linear_138_cast_fp16)[name = tensor<string, []>("op_1523_cast_fp16")];
            tensor<int32, [4]> var_1525 = const()[name = tensor<string, []>("op_1525"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1526_cast_fp16 = reshape(shape = var_1525, x = linear_139_cast_fp16)[name = tensor<string, []>("op_1526_cast_fp16")];
            tensor<int32, [4]> var_1528 = const()[name = tensor<string, []>("op_1528"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1529_cast_fp16 = reshape(shape = var_1528, x = linear_140_cast_fp16)[name = tensor<string, []>("op_1529_cast_fp16")];
            tensor<int32, [4]> value_states_95_perm_0 = const()[name = tensor<string, []>("value_states_95_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1532_transpose_x_0 = const()[name = tensor<string, []>("op_1532_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1532_transpose_y_0 = const()[name = tensor<string, []>("op_1532_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_127_perm_0 = const()[name = tensor<string, []>("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_128_perm_0 = const()[name = tensor<string, []>("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_1526_cast_fp16)[name = tensor<string, []>("transpose_148")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_127 = transpose(perm = transpose_127_perm_0, x = var_1523_cast_fp16)[name = tensor<string, []>("transpose_149")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1532_cast_fp16 = matmul(transpose_x = var_1532_transpose_x_0, transpose_y = var_1532_transpose_y_0, x = transpose_127, y = transpose_128)[name = tensor<string, []>("op_1532_cast_fp16")];
            tensor<fp16, []> var_1533_to_fp16 = const()[name = tensor<string, []>("op_1533_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_93_cast_fp16 = mul(x = var_1532_cast_fp16, y = var_1533_to_fp16)[name = tensor<string, []>("attn_weights_93_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1535_cast_fp16 = softmax(axis = var_11, x = attn_weights_93_cast_fp16)[name = tensor<string, []>("op_1535_cast_fp16")];
            tensor<bool, []> attn_output_93_transpose_x_0 = const()[name = tensor<string, []>("attn_output_93_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_93_transpose_y_0 = const()[name = tensor<string, []>("attn_output_93_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_95_cast_fp16 = transpose(perm = value_states_95_perm_0, x = var_1529_cast_fp16)[name = tensor<string, []>("transpose_150")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_93_cast_fp16 = matmul(transpose_x = attn_output_93_transpose_x_0, transpose_y = attn_output_93_transpose_y_0, x = var_1535_cast_fp16, y = value_states_95_cast_fp16)[name = tensor<string, []>("attn_output_93_cast_fp16")];
            tensor<int32, [4]> var_1539_perm_0 = const()[name = tensor<string, []>("op_1539_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1541 = const()[name = tensor<string, []>("op_1541"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1539_cast_fp16 = transpose(perm = var_1539_perm_0, x = attn_output_93_cast_fp16)[name = tensor<string, []>("transpose_147")];
            tensor<fp16, [1, 1024, 1152]> input_325_cast_fp16 = reshape(shape = var_1541, x = var_1539_cast_fp16)[name = tensor<string, []>("input_325_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_23_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709016192)))];
            tensor<fp16, [1152]> encoder_layers_23_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711670464)))];
            tensor<fp16, [1, 1024, 1152]> linear_141_cast_fp16 = linear(bias = encoder_layers_23_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_23_self_attn_out_proj_weight_to_fp16, x = input_325_cast_fp16)[name = tensor<string, []>("linear_141_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_327_cast_fp16 = add(x = input_321_cast_fp16, y = linear_141_cast_fp16)[name = tensor<string, []>("input_327_cast_fp16")];
            tensor<int32, [1]> input_329_axes_0 = const()[name = tensor<string, []>("input_329_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_23_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711672832)))];
            tensor<fp16, [1152]> encoder_layers_23_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711675200)))];
            tensor<fp16, [1, 1024, 1152]> input_329_cast_fp16 = layer_norm(axes = input_329_axes_0, beta = encoder_layers_23_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_23_layer_norm2_weight_to_fp16, x = input_327_cast_fp16)[name = tensor<string, []>("input_329_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_23_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(711677568)))];
            tensor<fp16, [4304]> encoder_layers_23_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721594048)))];
            tensor<fp16, [1, 1024, 4304]> linear_142_cast_fp16 = linear(bias = encoder_layers_23_mlp_fc1_bias_to_fp16, weight = encoder_layers_23_mlp_fc1_weight_to_fp16, x = input_329_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")];
            tensor<string, []> input_333_mode_0 = const()[name = tensor<string, []>("input_333_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_333_cast_fp16 = gelu(mode = input_333_mode_0, x = linear_142_cast_fp16)[name = tensor<string, []>("input_333_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_23_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721602752)))];
            tensor<fp16, [1152]> encoder_layers_23_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_23_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(731519232)))];
            tensor<fp16, [1, 1024, 1152]> linear_143_cast_fp16 = linear(bias = encoder_layers_23_mlp_fc2_bias_to_fp16, weight = encoder_layers_23_mlp_fc2_weight_to_fp16, x = input_333_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_335_cast_fp16 = add(x = input_327_cast_fp16, y = linear_143_cast_fp16)[name = tensor<string, []>("input_335_cast_fp16")];
            tensor<int32, [1]> hidden_states_145_axes_0 = const()[name = tensor<string, []>("hidden_states_145_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_24_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(731521600)))];
            tensor<fp16, [1152]> encoder_layers_24_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(731523968)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_145_cast_fp16 = layer_norm(axes = hidden_states_145_axes_0, beta = encoder_layers_24_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_24_layer_norm1_weight_to_fp16, x = input_335_cast_fp16)[name = tensor<string, []>("hidden_states_145_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(731526336)))];
            tensor<fp16, [1152]> encoder_layers_24_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734180608)))];
            tensor<fp16, [1, 1024, 1152]> linear_144_cast_fp16 = linear(bias = encoder_layers_24_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_q_proj_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(734182976)))];
            tensor<fp16, [1152]> encoder_layers_24_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736837248)))];
            tensor<fp16, [1, 1024, 1152]> linear_145_cast_fp16 = linear(bias = encoder_layers_24_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_k_proj_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(736839616)))];
            tensor<fp16, [1152]> encoder_layers_24_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(739493888)))];
            tensor<fp16, [1, 1024, 1152]> linear_146_cast_fp16 = linear(bias = encoder_layers_24_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_v_proj_weight_to_fp16, x = hidden_states_145_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")];
            tensor<int32, [4]> var_1584 = const()[name = tensor<string, []>("op_1584"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1585_cast_fp16 = reshape(shape = var_1584, x = linear_144_cast_fp16)[name = tensor<string, []>("op_1585_cast_fp16")];
            tensor<int32, [4]> var_1587 = const()[name = tensor<string, []>("op_1587"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1588_cast_fp16 = reshape(shape = var_1587, x = linear_145_cast_fp16)[name = tensor<string, []>("op_1588_cast_fp16")];
            tensor<int32, [4]> var_1590 = const()[name = tensor<string, []>("op_1590"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1591_cast_fp16 = reshape(shape = var_1590, x = linear_146_cast_fp16)[name = tensor<string, []>("op_1591_cast_fp16")];
            tensor<int32, [4]> value_states_99_perm_0 = const()[name = tensor<string, []>("value_states_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1594_transpose_x_0 = const()[name = tensor<string, []>("op_1594_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1594_transpose_y_0 = const()[name = tensor<string, []>("op_1594_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_129_perm_0 = const()[name = tensor<string, []>("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_130_perm_0 = const()[name = tensor<string, []>("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_1588_cast_fp16)[name = tensor<string, []>("transpose_144")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_129 = transpose(perm = transpose_129_perm_0, x = var_1585_cast_fp16)[name = tensor<string, []>("transpose_145")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1594_cast_fp16 = matmul(transpose_x = var_1594_transpose_x_0, transpose_y = var_1594_transpose_y_0, x = transpose_129, y = transpose_130)[name = tensor<string, []>("op_1594_cast_fp16")];
            tensor<fp16, []> var_1595_to_fp16 = const()[name = tensor<string, []>("op_1595_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_97_cast_fp16 = mul(x = var_1594_cast_fp16, y = var_1595_to_fp16)[name = tensor<string, []>("attn_weights_97_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1597_cast_fp16 = softmax(axis = var_11, x = attn_weights_97_cast_fp16)[name = tensor<string, []>("op_1597_cast_fp16")];
            tensor<bool, []> attn_output_97_transpose_x_0 = const()[name = tensor<string, []>("attn_output_97_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_97_transpose_y_0 = const()[name = tensor<string, []>("attn_output_97_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_99_cast_fp16 = transpose(perm = value_states_99_perm_0, x = var_1591_cast_fp16)[name = tensor<string, []>("transpose_146")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_97_cast_fp16 = matmul(transpose_x = attn_output_97_transpose_x_0, transpose_y = attn_output_97_transpose_y_0, x = var_1597_cast_fp16, y = value_states_99_cast_fp16)[name = tensor<string, []>("attn_output_97_cast_fp16")];
            tensor<int32, [4]> var_1601_perm_0 = const()[name = tensor<string, []>("op_1601_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1603 = const()[name = tensor<string, []>("op_1603"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1601_cast_fp16 = transpose(perm = var_1601_perm_0, x = attn_output_97_cast_fp16)[name = tensor<string, []>("transpose_143")];
            tensor<fp16, [1, 1024, 1152]> input_339_cast_fp16 = reshape(shape = var_1603, x = var_1601_cast_fp16)[name = tensor<string, []>("input_339_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_24_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(739496256)))];
            tensor<fp16, [1152]> encoder_layers_24_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(742150528)))];
            tensor<fp16, [1, 1024, 1152]> linear_147_cast_fp16 = linear(bias = encoder_layers_24_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_24_self_attn_out_proj_weight_to_fp16, x = input_339_cast_fp16)[name = tensor<string, []>("linear_147_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_341_cast_fp16 = add(x = input_335_cast_fp16, y = linear_147_cast_fp16)[name = tensor<string, []>("input_341_cast_fp16")];
            tensor<int32, [1]> input_343_axes_0 = const()[name = tensor<string, []>("input_343_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_24_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(742152896)))];
            tensor<fp16, [1152]> encoder_layers_24_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(742155264)))];
            tensor<fp16, [1, 1024, 1152]> input_343_cast_fp16 = layer_norm(axes = input_343_axes_0, beta = encoder_layers_24_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_24_layer_norm2_weight_to_fp16, x = input_341_cast_fp16)[name = tensor<string, []>("input_343_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_24_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(742157632)))];
            tensor<fp16, [4304]> encoder_layers_24_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(752074112)))];
            tensor<fp16, [1, 1024, 4304]> linear_148_cast_fp16 = linear(bias = encoder_layers_24_mlp_fc1_bias_to_fp16, weight = encoder_layers_24_mlp_fc1_weight_to_fp16, x = input_343_cast_fp16)[name = tensor<string, []>("linear_148_cast_fp16")];
            tensor<string, []> input_347_mode_0 = const()[name = tensor<string, []>("input_347_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_347_cast_fp16 = gelu(mode = input_347_mode_0, x = linear_148_cast_fp16)[name = tensor<string, []>("input_347_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_24_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(752082816)))];
            tensor<fp16, [1152]> encoder_layers_24_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_24_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(761999296)))];
            tensor<fp16, [1, 1024, 1152]> linear_149_cast_fp16 = linear(bias = encoder_layers_24_mlp_fc2_bias_to_fp16, weight = encoder_layers_24_mlp_fc2_weight_to_fp16, x = input_347_cast_fp16)[name = tensor<string, []>("linear_149_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_349_cast_fp16 = add(x = input_341_cast_fp16, y = linear_149_cast_fp16)[name = tensor<string, []>("input_349_cast_fp16")];
            tensor<int32, [1]> hidden_states_151_axes_0 = const()[name = tensor<string, []>("hidden_states_151_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_25_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762001664)))];
            tensor<fp16, [1152]> encoder_layers_25_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762004032)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_151_cast_fp16 = layer_norm(axes = hidden_states_151_axes_0, beta = encoder_layers_25_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_25_layer_norm1_weight_to_fp16, x = input_349_cast_fp16)[name = tensor<string, []>("hidden_states_151_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762006400)))];
            tensor<fp16, [1152]> encoder_layers_25_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764660672)))];
            tensor<fp16, [1, 1024, 1152]> linear_150_cast_fp16 = linear(bias = encoder_layers_25_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_q_proj_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor<string, []>("linear_150_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764663040)))];
            tensor<fp16, [1152]> encoder_layers_25_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767317312)))];
            tensor<fp16, [1, 1024, 1152]> linear_151_cast_fp16 = linear(bias = encoder_layers_25_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_k_proj_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(767319680)))];
            tensor<fp16, [1152]> encoder_layers_25_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(769973952)))];
            tensor<fp16, [1, 1024, 1152]> linear_152_cast_fp16 = linear(bias = encoder_layers_25_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_v_proj_weight_to_fp16, x = hidden_states_151_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")];
            tensor<int32, [4]> var_1646 = const()[name = tensor<string, []>("op_1646"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1647_cast_fp16 = reshape(shape = var_1646, x = linear_150_cast_fp16)[name = tensor<string, []>("op_1647_cast_fp16")];
            tensor<int32, [4]> var_1649 = const()[name = tensor<string, []>("op_1649"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1650_cast_fp16 = reshape(shape = var_1649, x = linear_151_cast_fp16)[name = tensor<string, []>("op_1650_cast_fp16")];
            tensor<int32, [4]> var_1652 = const()[name = tensor<string, []>("op_1652"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1653_cast_fp16 = reshape(shape = var_1652, x = linear_152_cast_fp16)[name = tensor<string, []>("op_1653_cast_fp16")];
            tensor<int32, [4]> value_states_103_perm_0 = const()[name = tensor<string, []>("value_states_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1656_transpose_x_0 = const()[name = tensor<string, []>("op_1656_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1656_transpose_y_0 = const()[name = tensor<string, []>("op_1656_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_132_perm_0 = const()[name = tensor<string, []>("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_1650_cast_fp16)[name = tensor<string, []>("transpose_140")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_131 = transpose(perm = transpose_131_perm_0, x = var_1647_cast_fp16)[name = tensor<string, []>("transpose_141")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1656_cast_fp16 = matmul(transpose_x = var_1656_transpose_x_0, transpose_y = var_1656_transpose_y_0, x = transpose_131, y = transpose_132)[name = tensor<string, []>("op_1656_cast_fp16")];
            tensor<fp16, []> var_1657_to_fp16 = const()[name = tensor<string, []>("op_1657_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_101_cast_fp16 = mul(x = var_1656_cast_fp16, y = var_1657_to_fp16)[name = tensor<string, []>("attn_weights_101_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1659_cast_fp16 = softmax(axis = var_11, x = attn_weights_101_cast_fp16)[name = tensor<string, []>("op_1659_cast_fp16")];
            tensor<bool, []> attn_output_101_transpose_x_0 = const()[name = tensor<string, []>("attn_output_101_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_101_transpose_y_0 = const()[name = tensor<string, []>("attn_output_101_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_103_cast_fp16 = transpose(perm = value_states_103_perm_0, x = var_1653_cast_fp16)[name = tensor<string, []>("transpose_142")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_101_cast_fp16 = matmul(transpose_x = attn_output_101_transpose_x_0, transpose_y = attn_output_101_transpose_y_0, x = var_1659_cast_fp16, y = value_states_103_cast_fp16)[name = tensor<string, []>("attn_output_101_cast_fp16")];
            tensor<int32, [4]> var_1663_perm_0 = const()[name = tensor<string, []>("op_1663_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1665 = const()[name = tensor<string, []>("op_1665"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1663_cast_fp16 = transpose(perm = var_1663_perm_0, x = attn_output_101_cast_fp16)[name = tensor<string, []>("transpose_139")];
            tensor<fp16, [1, 1024, 1152]> input_353_cast_fp16 = reshape(shape = var_1665, x = var_1663_cast_fp16)[name = tensor<string, []>("input_353_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_25_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(769976320)))];
            tensor<fp16, [1152]> encoder_layers_25_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772630592)))];
            tensor<fp16, [1, 1024, 1152]> linear_153_cast_fp16 = linear(bias = encoder_layers_25_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_25_self_attn_out_proj_weight_to_fp16, x = input_353_cast_fp16)[name = tensor<string, []>("linear_153_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_355_cast_fp16 = add(x = input_349_cast_fp16, y = linear_153_cast_fp16)[name = tensor<string, []>("input_355_cast_fp16")];
            tensor<int32, [1]> input_357_axes_0 = const()[name = tensor<string, []>("input_357_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_25_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772632960)))];
            tensor<fp16, [1152]> encoder_layers_25_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772635328)))];
            tensor<fp16, [1, 1024, 1152]> input_357_cast_fp16 = layer_norm(axes = input_357_axes_0, beta = encoder_layers_25_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_25_layer_norm2_weight_to_fp16, x = input_355_cast_fp16)[name = tensor<string, []>("input_357_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_25_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772637696)))];
            tensor<fp16, [4304]> encoder_layers_25_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782554176)))];
            tensor<fp16, [1, 1024, 4304]> linear_154_cast_fp16 = linear(bias = encoder_layers_25_mlp_fc1_bias_to_fp16, weight = encoder_layers_25_mlp_fc1_weight_to_fp16, x = input_357_cast_fp16)[name = tensor<string, []>("linear_154_cast_fp16")];
            tensor<string, []> input_361_mode_0 = const()[name = tensor<string, []>("input_361_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_361_cast_fp16 = gelu(mode = input_361_mode_0, x = linear_154_cast_fp16)[name = tensor<string, []>("input_361_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_25_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(782562880)))];
            tensor<fp16, [1152]> encoder_layers_25_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_25_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792479360)))];
            tensor<fp16, [1, 1024, 1152]> linear_155_cast_fp16 = linear(bias = encoder_layers_25_mlp_fc2_bias_to_fp16, weight = encoder_layers_25_mlp_fc2_weight_to_fp16, x = input_361_cast_fp16)[name = tensor<string, []>("linear_155_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_363_cast_fp16 = add(x = input_355_cast_fp16, y = linear_155_cast_fp16)[name = tensor<string, []>("input_363_cast_fp16")];
            tensor<int32, [1]> hidden_states_157_axes_0 = const()[name = tensor<string, []>("hidden_states_157_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_26_layer_norm1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_layer_norm1_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792481728)))];
            tensor<fp16, [1152]> encoder_layers_26_layer_norm1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_layer_norm1_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792484096)))];
            tensor<fp16, [1, 1024, 1152]> hidden_states_157_cast_fp16 = layer_norm(axes = hidden_states_157_axes_0, beta = encoder_layers_26_layer_norm1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_26_layer_norm1_weight_to_fp16, x = input_363_cast_fp16)[name = tensor<string, []>("hidden_states_157_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(792486464)))];
            tensor<fp16, [1152]> encoder_layers_26_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795140736)))];
            tensor<fp16, [1, 1024, 1152]> linear_156_cast_fp16 = linear(bias = encoder_layers_26_self_attn_q_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_q_proj_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor<string, []>("linear_156_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(795143104)))];
            tensor<fp16, [1152]> encoder_layers_26_self_attn_k_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(797797376)))];
            tensor<fp16, [1, 1024, 1152]> linear_157_cast_fp16 = linear(bias = encoder_layers_26_self_attn_k_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_k_proj_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor<string, []>("linear_157_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(797799744)))];
            tensor<fp16, [1152]> encoder_layers_26_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(800454016)))];
            tensor<fp16, [1, 1024, 1152]> linear_158_cast_fp16 = linear(bias = encoder_layers_26_self_attn_v_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_v_proj_weight_to_fp16, x = hidden_states_157_cast_fp16)[name = tensor<string, []>("linear_158_cast_fp16")];
            tensor<int32, [4]> var_1708 = const()[name = tensor<string, []>("op_1708"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1709_cast_fp16 = reshape(shape = var_1708, x = linear_156_cast_fp16)[name = tensor<string, []>("op_1709_cast_fp16")];
            tensor<int32, [4]> var_1711 = const()[name = tensor<string, []>("op_1711"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1712_cast_fp16 = reshape(shape = var_1711, x = linear_157_cast_fp16)[name = tensor<string, []>("op_1712_cast_fp16")];
            tensor<int32, [4]> var_1714 = const()[name = tensor<string, []>("op_1714"), val = tensor<int32, [4]>([1, 1024, 16, 72])];
            tensor<fp16, [1, 1024, 16, 72]> var_1715_cast_fp16 = reshape(shape = var_1714, x = linear_158_cast_fp16)[name = tensor<string, []>("op_1715_cast_fp16")];
            tensor<int32, [4]> value_states_perm_0 = const()[name = tensor<string, []>("value_states_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<bool, []> var_1718_transpose_x_0 = const()[name = tensor<string, []>("op_1718_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> var_1718_transpose_y_0 = const()[name = tensor<string, []>("op_1718_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_134_perm_0 = const()[name = tensor<string, []>("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 72, 1024]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_1712_cast_fp16)[name = tensor<string, []>("transpose_136")];
            tensor<fp16, [1, 16, 1024, 72]> transpose_133 = transpose(perm = transpose_133_perm_0, x = var_1709_cast_fp16)[name = tensor<string, []>("transpose_137")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1718_cast_fp16 = matmul(transpose_x = var_1718_transpose_x_0, transpose_y = var_1718_transpose_y_0, x = transpose_133, y = transpose_134)[name = tensor<string, []>("op_1718_cast_fp16")];
            tensor<fp16, []> var_1719_to_fp16 = const()[name = tensor<string, []>("op_1719_to_fp16"), val = tensor<fp16, []>(0x1.e2cp-4)];
            tensor<fp16, [1, 16, 1024, 1024]> attn_weights_105_cast_fp16 = mul(x = var_1718_cast_fp16, y = var_1719_to_fp16)[name = tensor<string, []>("attn_weights_105_cast_fp16")];
            tensor<fp16, [1, 16, 1024, 1024]> var_1721_cast_fp16 = softmax(axis = var_11, x = attn_weights_105_cast_fp16)[name = tensor<string, []>("op_1721_cast_fp16")];
            tensor<bool, []> attn_output_105_transpose_x_0 = const()[name = tensor<string, []>("attn_output_105_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> attn_output_105_transpose_y_0 = const()[name = tensor<string, []>("attn_output_105_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 1024, 72]> value_states_cast_fp16 = transpose(perm = value_states_perm_0, x = var_1715_cast_fp16)[name = tensor<string, []>("transpose_138")];
            tensor<fp16, [1, 16, 1024, 72]> attn_output_105_cast_fp16 = matmul(transpose_x = attn_output_105_transpose_x_0, transpose_y = attn_output_105_transpose_y_0, x = var_1721_cast_fp16, y = value_states_cast_fp16)[name = tensor<string, []>("attn_output_105_cast_fp16")];
            tensor<int32, [4]> var_1725_perm_0 = const()[name = tensor<string, []>("op_1725_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1727 = const()[name = tensor<string, []>("op_1727"), val = tensor<int32, [3]>([1, 1024, 1152])];
            tensor<fp16, [1, 1024, 16, 72]> var_1725_cast_fp16 = transpose(perm = var_1725_perm_0, x = attn_output_105_cast_fp16)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [1, 1024, 1152]> input_367_cast_fp16 = reshape(shape = var_1727, x = var_1725_cast_fp16)[name = tensor<string, []>("input_367_cast_fp16")];
            tensor<fp16, [1152, 1152]> encoder_layers_26_self_attn_out_proj_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [1152, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(800456384)))];
            tensor<fp16, [1152]> encoder_layers_26_self_attn_out_proj_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803110656)))];
            tensor<fp16, [1, 1024, 1152]> linear_159_cast_fp16 = linear(bias = encoder_layers_26_self_attn_out_proj_bias_to_fp16, weight = encoder_layers_26_self_attn_out_proj_weight_to_fp16, x = input_367_cast_fp16)[name = tensor<string, []>("linear_159_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_369_cast_fp16 = add(x = input_363_cast_fp16, y = linear_159_cast_fp16)[name = tensor<string, []>("input_369_cast_fp16")];
            tensor<int32, [1]> input_371_axes_0 = const()[name = tensor<string, []>("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> encoder_layers_26_layer_norm2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_layer_norm2_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803113024)))];
            tensor<fp16, [1152]> encoder_layers_26_layer_norm2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_layer_norm2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803115392)))];
            tensor<fp16, [1, 1024, 1152]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = encoder_layers_26_layer_norm2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_26_layer_norm2_weight_to_fp16, x = input_369_cast_fp16)[name = tensor<string, []>("input_371_cast_fp16")];
            tensor<fp16, [4304, 1152]> encoder_layers_26_mlp_fc1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_mlp_fc1_weight_to_fp16"), val = tensor<fp16, [4304, 1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(803117760)))];
            tensor<fp16, [4304]> encoder_layers_26_mlp_fc1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_mlp_fc1_bias_to_fp16"), val = tensor<fp16, [4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(813034240)))];
            tensor<fp16, [1, 1024, 4304]> linear_160_cast_fp16 = linear(bias = encoder_layers_26_mlp_fc1_bias_to_fp16, weight = encoder_layers_26_mlp_fc1_weight_to_fp16, x = input_371_cast_fp16)[name = tensor<string, []>("linear_160_cast_fp16")];
            tensor<string, []> input_375_mode_0 = const()[name = tensor<string, []>("input_375_mode_0"), val = tensor<string, []>("TANH_APPROXIMATION")];
            tensor<fp16, [1, 1024, 4304]> input_375_cast_fp16 = gelu(mode = input_375_mode_0, x = linear_160_cast_fp16)[name = tensor<string, []>("input_375_cast_fp16")];
            tensor<fp16, [1152, 4304]> encoder_layers_26_mlp_fc2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_mlp_fc2_weight_to_fp16"), val = tensor<fp16, [1152, 4304]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(813042944)))];
            tensor<fp16, [1152]> encoder_layers_26_mlp_fc2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_26_mlp_fc2_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(822959424)))];
            tensor<fp16, [1, 1024, 1152]> linear_161_cast_fp16 = linear(bias = encoder_layers_26_mlp_fc2_bias_to_fp16, weight = encoder_layers_26_mlp_fc2_weight_to_fp16, x = input_375_cast_fp16)[name = tensor<string, []>("linear_161_cast_fp16")];
            tensor<fp16, [1, 1024, 1152]> input_cast_fp16 = add(x = input_369_cast_fp16, y = linear_161_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
            tensor<int32, [1]> var_1753_axes_0 = const()[name = tensor<string, []>("op_1753_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1152]> post_layernorm_weight_to_fp16 = const()[name = tensor<string, []>("post_layernorm_weight_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(822961792)))];
            tensor<fp16, [1152]> post_layernorm_bias_to_fp16 = const()[name = tensor<string, []>("post_layernorm_bias_to_fp16"), val = tensor<fp16, [1152]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(822964160)))];
            tensor<fp16, []> var_1748_to_fp16 = const()[name = tensor<string, []>("op_1748_to_fp16"), val = tensor<fp16, []>(0x1.1p-20)];
            tensor<fp16, [1, 1024, 1152]> var_1753_cast_fp16 = layer_norm(axes = var_1753_axes_0, beta = post_layernorm_bias_to_fp16, epsilon = var_1748_to_fp16, gamma = post_layernorm_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("op_1753_cast_fp16")];
            tensor<string, []> var_1753_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1753_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
            tensor<fp32, [1, 1024, 1152]> output = cast(dtype = var_1753_cast_fp16_to_fp32_dtype_0, x = var_1753_cast_fp16)[name = tensor<string, []>("cast_135")];
        } -> (output);
}