Instructions to use RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8 with Transformers:

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

pipe = pipeline("text-generation", model="RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8")
model = AutoModelForCausalLM.from_pretrained("RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8

SGLang

How to use RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8 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 "RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8" \
    --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": "RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8",
		"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 "RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8" \
        --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": "RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8 with Docker Model Runner:
```
docker model run hf.co/RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8
```

RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8

This is a quantized version of deepseek-ai/DeepSeek-V4-Pro with MoE layers quantized to NVFP4 and attention layers quantized to FP8 block

Usage

This model is intended for deployment with vLLM and requires the following branch: https://github.com/vllm-project/vllm/pull/41276. You can serve the model using

vllm serve RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8-BLOCK --tensor_parallel_size 8 --kv_cache_dtype=fp8

Creation Process

This model was created using LLM Compressor. The example script can be found in examples/quantizing_moe/deepseek_v4_pro_example.py [DSV4] DeepSeekV4 Pro. Quantizing the model with data parallelism and 6xA100 takes about 3 hours.

Evaluation

Benchmark	`deepseek-ai/DeepSeek-V4-Pro-Base`	`deepseek-ai/DeepSeek-V4-Pro`	`RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8`
GPQA		90.1	0.93 (380/792 samples)
GSM8K	91.1	92.6	91.0

Downloads last month: 276

Safetensors

Model size

894B params

Tensor type

I64

F32

BF16

F8_E4M3

Model tree for RedHatAI/DeepSeek-V4-Pro-NVFP4-FP8

Base model

deepseek-ai/DeepSeek-V4-Pro

Quantized

(23)

this model