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

Libraries

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

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

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

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

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

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

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

Use Docker

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

SGLang

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

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

krishnateja95 commited on Jun 26

Commit

ae58ee7

verified ·

1 Parent(s): 1f85199

Add accuracy recovery summary table with evaluation results

Browse files

Files changed (1) hide show

README.md +16 -0

README.md CHANGED Viewed

@@ -69,3 +69,19 @@ uv pip install git+https://github.com/vllm-project/vllm.git@refs/pull/41276/head
 uv pip install tilelang==0.1.10 apache-tvm-ffi==0.1.10
 ```

 uv pip install tilelang==0.1.10 apache-tvm-ffi==0.1.10
 ```
+## Accuracy Recovery Summary
+Evaluation performed on 8×B200 GPUs using [vLLM](https://github.com/vllm-project/vllm) with FP8 KV cache.
+Scores are averaged across multiple seeds (3 seeds for most benchmarks, 8 for AIME 2025).
+Instruct benchmarks run with reasoning OFF (nonthinking mode); Reasoning and Coding benchmarks run with reasoning ON (thinking mode).
+| Category | Benchmark | deepseek-ai/<br>DeepSeek-V4-Flash | RedHatAI/<br>DeepSeek-V4-Flash-NVFP4-FP8<br>(this model) | Recovery |
+|:---------|:----------|:-----------------------------:|:--------------------------------------------------:|:--------:|
+| Instruct | MMLU-CoT (5-shot) | 86.10 | 78.39 | 91.05% |
+| Instruct | GSM8K Platinum (5-shot) | 96.99 | 94.07 | 96.99% |
+| Instruct | MATH-500 | 91.93 | 89.73 | 97.61% |
+| Reasoning | GSM8K Platinum (0-shot) | 95.62 | 94.13 | 98.44% |
+| Reasoning | MATH-500 | 91.67 | 89.87 | 98.04% |
+| Reasoning | AIME 2025 | 52.92 | 72.08 | 136.22% |
+| Coding | LCB CodeGen v6 | 51.81 | 48.00 | 92.65% |