Instructions to use NobodyExistsOnTheInternet/bigseek_3p2t with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use NobodyExistsOnTheInternet/bigseek_3p2t with Transformers:

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

pipe = pipeline("text-generation", model="NobodyExistsOnTheInternet/bigseek_3p2t")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("NobodyExistsOnTheInternet/bigseek_3p2t")
model = AutoModelForCausalLM.from_pretrained("NobodyExistsOnTheInternet/bigseek_3p2t", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use NobodyExistsOnTheInternet/bigseek_3p2t with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "NobodyExistsOnTheInternet/bigseek_3p2t"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NobodyExistsOnTheInternet/bigseek_3p2t",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/NobodyExistsOnTheInternet/bigseek_3p2t

SGLang

How to use NobodyExistsOnTheInternet/bigseek_3p2t 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 "NobodyExistsOnTheInternet/bigseek_3p2t" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NobodyExistsOnTheInternet/bigseek_3p2t",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "NobodyExistsOnTheInternet/bigseek_3p2t" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "NobodyExistsOnTheInternet/bigseek_3p2t",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use NobodyExistsOnTheInternet/bigseek_3p2t with Docker Model Runner:
```
docker model run hf.co/NobodyExistsOnTheInternet/bigseek_3p2t
```

bigseek_3p2t / config.json

NobodyExistsOnTheInternet

Add files using upload-large-folder tool

c6f0223 verified 3 months ago

Raw

History Blame

8.55 kB

	{
	"architectures": [
	"DeepseekV4ForCausalLM"
	],
	"attention_bias": false,
	"attention_dropout": 0.0,
	"bos_token_id": 0,
	"compress_ratios": [
	128,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	128,
	4,
	0
	],
	"compress_rope_theta": 160000,
	"eos_token_id": 1,
	"expert_dtype": "fp8",
	"hc_eps": 1e-06,
	"hc_mult": 4,
	"hc_sinkhorn_iters": 20,
	"head_dim": 512,
	"hidden_act": "silu",
	"hidden_size": 7168,
	"index_head_dim": 128,
	"index_n_heads": 64,
	"index_topk": 1024,
	"initializer_range": 0.02,
	"max_position_embeddings": 1048576,
	"mergehell_latest_indexer_patch": {
	"index_head_dim": 128,
	"index_n_heads": 64,
	"note": "Sparse attention indexer tensors were replaced with native DeepSeek-V4-Pro instruct indexer tensors. The merged checkpoint keeps widened main attention heads, but the indexer now uses the trained 64-head instruct retrieval policy.",
	"patch": "use_native_instruct_sparse_indexer",
	"patch_shard": "model-instruct-indexer64-patch.safetensors",
	"patched_at_utc": "2026-05-29T05:36:52Z",
	"previous_index_n_heads": 128,
	"source_layer_for_target_layer": {
	"10": 10,
	"12": 12,
	"14": 14,
	"16": 16,
	"18": 18,
	"2": 2,
	"20": 20,
	"22": 22,
	"24": 24,
	"26": 26,
	"28": 28,
	"30": 30,
	"32": 32,
	"34": 34,
	"36": 36,
	"38": 38,
	"4": 4,
	"40": 40,
	"42": 42,
	"44": 44,
	"46": 46,
	"48": 48,
	"50": 50,
	"52": 52,
	"54": 54,
	"56": 56,
	"58": 58,
	"6": 6,
	"60": 60,
	"8": 8
	},
	"source_model": "/home/shared/models/DeepSeek-V4-Pro",
	"target_layer_count": 30,
	"target_layers": [
	2,
	4,
	6,
	8,
	10,
	12,
	14,
	16,
	18,
	20,
	22,
	24,
	26,
	28,
	30,
	32,
	34,
	36,
	38,
	40,
	42,
	44,
	46,
	48,
	50,
	52,
	54,
	56,
	58,
	60
	],
	"tensor_families": [
	"attn.indexer.wq_b.weight",
	"attn.indexer.wq_b.scale",
	"attn.indexer.weights_proj.weight",
	"attn.indexer.compressor.wkv.weight",
	"attn.indexer.compressor.wgate.weight",
	"attn.indexer.compressor.norm.weight",
	"attn.indexer.compressor.ape"
	]
	},
	"mergehell_patch_history": [
	{
	"index_head_dim": 128,
	"index_n_heads": 64,
	"note": "Sparse attention indexer tensors were replaced with native DeepSeek-V4-Pro instruct indexer tensors. The merged checkpoint keeps widened main attention heads, but the indexer now uses the trained 64-head instruct retrieval policy.",
	"patch": "use_native_instruct_sparse_indexer",
	"patch_shard": "model-instruct-indexer64-patch.safetensors",
	"patched_at_utc": "2026-05-29T05:36:52Z",
	"previous_index_n_heads": 128,
	"source_layer_for_target_layer": {
	"10": 10,
	"12": 12,
	"14": 14,
	"16": 16,
	"18": 18,
	"2": 2,
	"20": 20,
	"22": 22,
	"24": 24,
	"26": 26,
	"28": 28,
	"30": 30,
	"32": 32,
	"34": 34,
	"36": 36,
	"38": 38,
	"4": 4,
	"40": 40,
	"42": 42,
	"44": 44,
	"46": 46,
	"48": 48,
	"50": 50,
	"52": 52,
	"54": 54,
	"56": 56,
	"58": 58,
	"6": 6,
	"60": 60,
	"8": 8
	},
	"source_model": "/home/shared/models/DeepSeek-V4-Pro",
	"target_layer_count": 30,
	"target_layers": [
	2,
	4,
	6,
	8,
	10,
	12,
	14,
	16,
	18,
	20,
	22,
	24,
	26,
	28,
	30,
	32,
	34,
	36,
	38,
	40,
	42,
	44,
	46,
	48,
	50,
	52,
	54,
	56,
	58,
	60
	],
	"tensor_families": [
	"attn.indexer.wq_b.weight",
	"attn.indexer.wq_b.scale",
	"attn.indexer.weights_proj.weight",
	"attn.indexer.compressor.wkv.weight",
	"attn.indexer.compressor.wgate.weight",
	"attn.indexer.compressor.norm.weight",
	"attn.indexer.compressor.ape"
	]
	}
	],
	"model_type": "deepseek_v4",
	"moe_intermediate_size": 3072,
	"n_routed_experts": 768,
	"n_shared_experts": 2,
	"norm_topk_prob": true,
	"num_attention_heads": 256,
	"num_experts_per_tok": 12,
	"num_hash_layers": 3,
	"num_hidden_layers": 61,
	"num_key_value_heads": 1,
	"num_nextn_predict_layers": 1,
	"o_groups": 32,
	"o_lora_rank": 1024,
	"q_lora_rank": 1536,
	"qk_rope_head_dim": 64,
	"quantization_config": {
	"activation_scheme": "dynamic",
	"fmt": "e4m3",
	"ignored_layers": [
	"model.layers.0.attn.fused_wqa_wkv",
	"model.layers.1.attn.fused_wqa_wkv",
	"model.layers.2.attn.fused_wqa_wkv",
	"model.layers.3.attn.fused_wqa_wkv",
	"model.layers.4.attn.fused_wqa_wkv",
	"model.layers.5.attn.fused_wqa_wkv",
	"model.layers.6.attn.fused_wqa_wkv",
	"model.layers.7.attn.fused_wqa_wkv",
	"model.layers.8.attn.fused_wqa_wkv",
	"model.layers.9.attn.fused_wqa_wkv",
	"model.layers.10.attn.fused_wqa_wkv",
	"model.layers.11.attn.fused_wqa_wkv",
	"model.layers.12.attn.fused_wqa_wkv",
	"model.layers.13.attn.fused_wqa_wkv",
	"model.layers.14.attn.fused_wqa_wkv",
	"model.layers.15.attn.fused_wqa_wkv",
	"model.layers.16.attn.fused_wqa_wkv",
	"model.layers.17.attn.fused_wqa_wkv",
	"model.layers.18.attn.fused_wqa_wkv",
	"model.layers.19.attn.fused_wqa_wkv",
	"model.layers.20.attn.fused_wqa_wkv",
	"model.layers.21.attn.fused_wqa_wkv",
	"model.layers.22.attn.fused_wqa_wkv",
	"model.layers.23.attn.fused_wqa_wkv",
	"model.layers.24.attn.fused_wqa_wkv",
	"model.layers.25.attn.fused_wqa_wkv",
	"model.layers.26.attn.fused_wqa_wkv",
	"model.layers.27.attn.fused_wqa_wkv",
	"model.layers.28.attn.fused_wqa_wkv",
	"model.layers.29.attn.fused_wqa_wkv",
	"model.layers.30.attn.fused_wqa_wkv",
	"model.layers.31.attn.fused_wqa_wkv",
	"model.layers.32.attn.fused_wqa_wkv",
	"model.layers.33.attn.fused_wqa_wkv",
	"model.layers.34.attn.fused_wqa_wkv",
	"model.layers.35.attn.fused_wqa_wkv",
	"model.layers.36.attn.fused_wqa_wkv",
	"model.layers.37.attn.fused_wqa_wkv",
	"model.layers.38.attn.fused_wqa_wkv",
	"model.layers.39.attn.fused_wqa_wkv",
	"model.layers.40.attn.fused_wqa_wkv",
	"model.layers.41.attn.fused_wqa_wkv",
	"model.layers.42.attn.fused_wqa_wkv",
	"model.layers.43.attn.fused_wqa_wkv",
	"model.layers.44.attn.fused_wqa_wkv",
	"model.layers.45.attn.fused_wqa_wkv",
	"model.layers.46.attn.fused_wqa_wkv",
	"model.layers.47.attn.fused_wqa_wkv",
	"model.layers.48.attn.fused_wqa_wkv",
	"model.layers.49.attn.fused_wqa_wkv",
	"model.layers.50.attn.fused_wqa_wkv",
	"model.layers.51.attn.fused_wqa_wkv",
	"model.layers.52.attn.fused_wqa_wkv",
	"model.layers.53.attn.fused_wqa_wkv",
	"model.layers.54.attn.fused_wqa_wkv",
	"model.layers.55.attn.fused_wqa_wkv",
	"model.layers.56.attn.fused_wqa_wkv",
	"model.layers.57.attn.fused_wqa_wkv",
	"model.layers.58.attn.fused_wqa_wkv",
	"model.layers.59.attn.fused_wqa_wkv",
	"model.layers.60.attn.fused_wqa_wkv",
	"model.mtp.0.attn.fused_wqa_wkv"
	],
	"quant_method": "fp8",
	"scale_fmt": "ue8m0",
	"weight_block_size": [
	128,
	128
	]
	},
	"rms_norm_eps": 1e-06,
	"rope_scaling": {
	"beta_fast": 32,
	"beta_slow": 1,
	"factor": 16,
	"original_max_position_embeddings": 65536,
	"type": "yarn"
	},
	"rope_theta": 10000,
	"routed_scaling_factor": 2.5,
	"scoring_func": "sqrtsoftplus",
	"sliding_window": 128,
	"swiglu_limit": 10.0,
	"tie_word_embeddings": false,
	"topk_method": "noaux_tc",
	"torch_dtype": "bfloat16",
	"transformers_version": "4.57.1",
	"use_cache": true,
	"vocab_size": 129280
	}