Instructions to use DataSnake/Muse-12B-NVFP4-4over6 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DataSnake/Muse-12B-NVFP4-4over6 with Transformers:

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

pipe = pipeline("text-generation", model="DataSnake/Muse-12B-NVFP4-4over6")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("DataSnake/Muse-12B-NVFP4-4over6")
model = AutoModelForCausalLM.from_pretrained("DataSnake/Muse-12B-NVFP4-4over6")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.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(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use DataSnake/Muse-12B-NVFP4-4over6 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DataSnake/Muse-12B-NVFP4-4over6"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DataSnake/Muse-12B-NVFP4-4over6",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/DataSnake/Muse-12B-NVFP4-4over6

SGLang

How to use DataSnake/Muse-12B-NVFP4-4over6 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 "DataSnake/Muse-12B-NVFP4-4over6" \
    --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": "DataSnake/Muse-12B-NVFP4-4over6",
		"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 "DataSnake/Muse-12B-NVFP4-4over6" \
        --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": "DataSnake/Muse-12B-NVFP4-4over6",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use DataSnake/Muse-12B-NVFP4-4over6 with Docker Model Runner:
```
docker model run hf.co/DataSnake/Muse-12B-NVFP4-4over6
```

DataSnake commited on Feb 23

Commit

ed6b1a9

verified ·

1 Parent(s): c924e31

Upload 2 files

Browse files

Updated to use the correct files, which set `norm` to `2.0` for the observer to properly emulate MSE selection.

Files changed (2) hide show

config.json +2 -1
recipe.yaml +1 -1

config.json CHANGED Viewed

@@ -48,7 +48,8 @@
           "observer": "memoryless_mse",
           "observer_kwargs": {
             "grid": -2.0,
-            "maxshrink": -1.0
           },
           "scale_dtype": "torch.float8_e4m3fn",
           "strategy": "tensor_group",

           "observer": "memoryless_mse",
           "observer_kwargs": {
             "grid": -2.0,
+            "maxshrink": -1.0,
+            "norm": 2.0
           },
           "scale_dtype": "torch.float8_e4m3fn",
           "strategy": "tensor_group",

recipe.yaml CHANGED Viewed

@@ -16,7 +16,7 @@ default_stage:
             scale_dtype: torch.float8_e4m3fn
             zp_dtype: null
             observer: memoryless_mse
-            observer_kwargs: {maxshrink: -1.0, grid: -2.0}
           input_activations:
             num_bits: 4
             type: float

             scale_dtype: torch.float8_e4m3fn
             zp_dtype: null
             observer: memoryless_mse
+            observer_kwargs: {maxshrink: -1.0, grid: -2.0, norm: 2.0}
           input_activations:
             num_bits: 4
             type: float