Instructions to use atbender/Qwen3.5-REAP-212B-A17B-W4A16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use atbender/Qwen3.5-REAP-212B-A17B-W4A16 with Transformers:

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

pipe = pipeline("text-generation", model="atbender/Qwen3.5-REAP-212B-A17B-W4A16")
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 AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("atbender/Qwen3.5-REAP-212B-A17B-W4A16")
model = AutoModelForMultimodalLM.from_pretrained("atbender/Qwen3.5-REAP-212B-A17B-W4A16", device_map="auto")
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?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use atbender/Qwen3.5-REAP-212B-A17B-W4A16 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "atbender/Qwen3.5-REAP-212B-A17B-W4A16"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "atbender/Qwen3.5-REAP-212B-A17B-W4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/atbender/Qwen3.5-REAP-212B-A17B-W4A16

SGLang

How to use atbender/Qwen3.5-REAP-212B-A17B-W4A16 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 "atbender/Qwen3.5-REAP-212B-A17B-W4A16" \
    --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": "atbender/Qwen3.5-REAP-212B-A17B-W4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

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 "atbender/Qwen3.5-REAP-212B-A17B-W4A16" \
        --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": "atbender/Qwen3.5-REAP-212B-A17B-W4A16",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use atbender/Qwen3.5-REAP-212B-A17B-W4A16 with Docker Model Runner:
```
docker model run hf.co/atbender/Qwen3.5-REAP-212B-A17B-W4A16
```

"vLLM Launch Failed - Kernel Incompatibility with RTX 3090"

by steppi - opened Mar 2

Discussion

steppi

Mar 2

Hi! I'm trying to run this model on 8x RTX 3090 (24GB each) with vLLM 0.16.1, but getting: "AllSparkLinearKernel cannot implement due to: For Ampere GPU, AllSpark does not support group_size = 128". Full error: All kernels require Compute 9.0, RTX 3090 has 8.6. The model creator shared a config that uses TP=8, EP, etc. - but it seems to require Hopper GPUs. Has anyone successfully run this on RTX 3090/Ampere? What vLLM version/commit works? Thanks!

Docker version:
docker run -d
--gpus all
-p 8000:8000
--ipc=host
-v ~/.cache/huggingface:/root/.cache/huggingface
--name vllm-qwen35-reap
vllm/vllm-openai:nightly
atbender/Qwen3.5-REAP-212B-A17B-W4A16
--tensor-parallel-size 8
--pipeline-parallel-size 1
--max-model-len 200000
--gpu-memory-utilization 0.9
--kv-cache-dtype fp8_e4m3
--max-num-seqs 8
--trust-remote-code
--tool-call-parser qwen3_coder
--reasoning-parser qwen3
--enable-auto-tool-choice
--enable-expert-parallel
--enable-chunked-prefill
--served-model-name qwen35-reap-212b

Native vLLM version:
vllm serve atbender/Qwen3.5-REAP-212B-A17B-W4A16
--tensor-parallel-size 8
--pipeline-parallel-size 1
--max-model-len 200000
--gpu-memory-utilization 0.9
--kv-cache-dtype fp8_e4m3
--max-num-seqs 8
--trust-remote-code
--tool-call-parser qwen3_coder
--reasoning-parser qwen3
--enable-auto-tool-choice
--enable-expert-parallel
--enable-chunked-prefill
--served-model-name qwen35-reap-212b

0xSero

Mar 2

"name": "Qwen3.5-REAP-212B-A17B W4A16 (vLLM)",
"model_path": "/mnt/llm_models/Qwen3.5-REAP-212B-A17B-W4A16",
"backend": "vllm",
"tensor_parallel_size": 8,
"pipeline_parallel_size": 1,
"max_model_len": 200000,
"gpu_memory_utilization": 0.9,
"kv_cache_dtype": "fp8_e4m3",
"max_num_seqs": 8,
"trust_remote_code": true,
"tool_call_parser": "qwen3_coder",
"reasoning_parser": "qwen3",
"enable_auto_tool_choice": true,
"host": "0.0.0.0",
"port": 8000,
"served_model_name": "qwen35-reap-212b",
"python_path": "/opt/venvs/active/vllm-latest/bin/python",
"extra_args": {
"enable_expert_parallel": true,
"enable_chunked_prefill": true,
"default_chat_template_kwargs": {
"enable_thinking": true
}
},
"thinking_mode": "conservative"

mwalol

Mar 3

0xSero I’ve seen low-effort replies before, but this sets a new standard for disappointing. Did you even try?

0xSero

Mar 3

0xSero I’ve seen low-effort replies before, but this sets a new standard for disappointing. Did you even try?

lol what? you're not owed a reply, I have been using this model and have contributed to the production of it, figure it out sir.

mwalol

Mar 5

your arrogance doesn't have any value here. I've already figured it out, so you can spare the condescension. Contributing to the model doesn't make you superior to the people using it.

steppi

Mar 5

Calm down, guys... I just asked for some more info because I've already wasted three evenings trying to get this model working. It would be really helpful for me to get it working, but I just can't. Can you point me to a special version of VLLM you used? I always get that error.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment