Instructions to use stepfun-ai/Step-3.7-Flash-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use stepfun-ai/Step-3.7-Flash-NVFP4 with Transformers:

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

pipe = pipeline("image-text-to-text", model="stepfun-ai/Step-3.7-Flash-NVFP4", 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 AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("stepfun-ai/Step-3.7-Flash-NVFP4", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use stepfun-ai/Step-3.7-Flash-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "stepfun-ai/Step-3.7-Flash-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "stepfun-ai/Step-3.7-Flash-NVFP4",
		"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/stepfun-ai/Step-3.7-Flash-NVFP4

SGLang

How to use stepfun-ai/Step-3.7-Flash-NVFP4 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 "stepfun-ai/Step-3.7-Flash-NVFP4" \
    --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": "stepfun-ai/Step-3.7-Flash-NVFP4",
		"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 "stepfun-ai/Step-3.7-Flash-NVFP4" \
        --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": "stepfun-ai/Step-3.7-Flash-NVFP4",
		"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"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use stepfun-ai/Step-3.7-Flash-NVFP4 with Docker Model Runner:
```
docker model run hf.co/stepfun-ai/Step-3.7-Flash-NVFP4
```

Step-3.7-Flash-NVFP4 / hf_quant_config.json

huangyu-nv

Upload Step3.7 NVFP4 checkpoint from step3p7-nvfp4-moe-only-mix_nvidia-calib4096-kvfp8

d8c9009 verified 2 days ago

raw

history blame contribute delete

7.71 kB

	{
	"producer": {
	"name": "modelopt",
	"version": "0.45.0.dev37+g3ad4f4f09.d20260524"
	},
	"quantization": {
	"quant_algo": "NVFP4",
	"kv_cache_quant_algo": "FP8",
	"group_size": 16,
	"exclude_modules": [
	"lm_head",
	"model.language_model.layers.0*",
	"model.language_model.layers.1.*",
	"model.language_model.layers.10.moe.gate",
	"model.language_model.layers.10.self_attn*",
	"model.language_model.layers.10.share_expert*",
	"model.language_model.layers.11.moe.gate",
	"model.language_model.layers.11.self_attn*",
	"model.language_model.layers.11.share_expert*",
	"model.language_model.layers.12.moe.gate",
	"model.language_model.layers.12.self_attn*",
	"model.language_model.layers.12.share_expert*",
	"model.language_model.layers.13.moe.gate",
	"model.language_model.layers.13.self_attn*",
	"model.language_model.layers.13.share_expert*",
	"model.language_model.layers.14.moe.gate",
	"model.language_model.layers.14.self_attn*",
	"model.language_model.layers.14.share_expert*",
	"model.language_model.layers.15.moe.gate",
	"model.language_model.layers.15.self_attn*",
	"model.language_model.layers.15.share_expert*",
	"model.language_model.layers.16.moe.gate",
	"model.language_model.layers.16.self_attn*",
	"model.language_model.layers.16.share_expert*",
	"model.language_model.layers.17.moe.gate",
	"model.language_model.layers.17.self_attn*",
	"model.language_model.layers.17.share_expert*",
	"model.language_model.layers.18.moe.gate",
	"model.language_model.layers.18.self_attn*",
	"model.language_model.layers.18.share_expert*",
	"model.language_model.layers.19.moe.gate",
	"model.language_model.layers.19.self_attn*",
	"model.language_model.layers.19.share_expert*",
	"model.language_model.layers.2.*",
	"model.language_model.layers.20.moe.gate",
	"model.language_model.layers.20.self_attn*",
	"model.language_model.layers.20.share_expert*",
	"model.language_model.layers.21.moe.gate",
	"model.language_model.layers.21.self_attn*",
	"model.language_model.layers.21.share_expert*",
	"model.language_model.layers.22.moe.gate",
	"model.language_model.layers.22.self_attn*",
	"model.language_model.layers.22.share_expert*",
	"model.language_model.layers.23.moe.gate",
	"model.language_model.layers.23.self_attn*",
	"model.language_model.layers.23.share_expert*",
	"model.language_model.layers.24.moe.gate",
	"model.language_model.layers.24.self_attn*",
	"model.language_model.layers.24.share_expert*",
	"model.language_model.layers.25.moe.gate",
	"model.language_model.layers.25.self_attn*",
	"model.language_model.layers.25.share_expert*",
	"model.language_model.layers.26.moe.gate",
	"model.language_model.layers.26.self_attn*",
	"model.language_model.layers.26.share_expert*",
	"model.language_model.layers.27.moe.gate",
	"model.language_model.layers.27.self_attn*",
	"model.language_model.layers.27.share_expert*",
	"model.language_model.layers.28.moe.gate",
	"model.language_model.layers.28.self_attn*",
	"model.language_model.layers.28.share_expert*",
	"model.language_model.layers.29.moe.gate",
	"model.language_model.layers.29.self_attn*",
	"model.language_model.layers.29.share_expert*",
	"model.language_model.layers.3.moe.gate",
	"model.language_model.layers.3.self_attn*",
	"model.language_model.layers.3.share_expert*",
	"model.language_model.layers.30.moe.gate",
	"model.language_model.layers.30.self_attn*",
	"model.language_model.layers.30.share_expert*",
	"model.language_model.layers.31.moe.gate",
	"model.language_model.layers.31.self_attn*",
	"model.language_model.layers.31.share_expert*",
	"model.language_model.layers.32.moe.gate",
	"model.language_model.layers.32.self_attn*",
	"model.language_model.layers.32.share_expert*",
	"model.language_model.layers.33.moe.gate",
	"model.language_model.layers.33.self_attn*",
	"model.language_model.layers.33.share_expert*",
	"model.language_model.layers.34.moe.gate",
	"model.language_model.layers.34.self_attn*",
	"model.language_model.layers.34.share_expert*",
	"model.language_model.layers.35.moe.gate",
	"model.language_model.layers.35.self_attn*",
	"model.language_model.layers.35.share_expert*",
	"model.language_model.layers.36.moe.gate",
	"model.language_model.layers.36.self_attn*",
	"model.language_model.layers.36.share_expert*",
	"model.language_model.layers.37.moe.gate",
	"model.language_model.layers.37.self_attn*",
	"model.language_model.layers.37.share_expert*",
	"model.language_model.layers.38.moe.gate",
	"model.language_model.layers.38.self_attn*",
	"model.language_model.layers.38.share_expert*",
	"model.language_model.layers.39.moe.gate",
	"model.language_model.layers.39.self_attn*",
	"model.language_model.layers.39.share_expert*",
	"model.language_model.layers.4.moe.gate",
	"model.language_model.layers.4.self_attn*",
	"model.language_model.layers.4.share_expert*",
	"model.language_model.layers.40.moe.gate",
	"model.language_model.layers.40.self_attn*",
	"model.language_model.layers.40.share_expert*",
	"model.language_model.layers.41.moe.gate",
	"model.language_model.layers.41.self_attn*",
	"model.language_model.layers.41.share_expert*",
	"model.language_model.layers.42.moe.gate",
	"model.language_model.layers.42.self_attn*",
	"model.language_model.layers.42.share_expert*",
	"model.language_model.layers.43.moe.gate",
	"model.language_model.layers.43.self_attn*",
	"model.language_model.layers.43.share_expert*",
	"model.language_model.layers.44.moe.gate",
	"model.language_model.layers.44.self_attn*",
	"model.language_model.layers.44.share_expert*",
	"model.language_model.layers.5.moe.gate",
	"model.language_model.layers.5.self_attn*",
	"model.language_model.layers.5.share_expert*",
	"model.language_model.layers.6.moe.gate",
	"model.language_model.layers.6.self_attn*",
	"model.language_model.layers.6.share_expert*",
	"model.language_model.layers.7.moe.gate",
	"model.language_model.layers.7.self_attn*",
	"model.language_model.layers.7.share_expert*",
	"model.language_model.layers.8.moe.gate",
	"model.language_model.layers.8.self_attn*",
	"model.language_model.layers.8.share_expert*",
	"model.language_model.layers.9.moe.gate",
	"model.language_model.layers.9.self_attn*",
	"model.language_model.layers.9.share_expert*",
	"model.vision_model*",
	"model.vit_large_projector"
	]
	}
	}