deepseek-ai
/

DeepSeek-V4-Flash-0731

Text Generation

8-bit precision

Model card Files Files and versions

Instructions to use deepseek-ai/DeepSeek-V4-Flash-0731 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use deepseek-ai/DeepSeek-V4-Flash-0731 with Transformers:

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

pipe = pipeline("text-generation", model="deepseek-ai/DeepSeek-V4-Flash-0731")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V4-Flash-0731")
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-V4-Flash-0731", device_map="auto")

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use deepseek-ai/DeepSeek-V4-Flash-0731 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "deepseek-ai/DeepSeek-V4-Flash-0731"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "deepseek-ai/DeepSeek-V4-Flash-0731",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/deepseek-ai/DeepSeek-V4-Flash-0731

How to use deepseek-ai/DeepSeek-V4-Flash-0731 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 "deepseek-ai/DeepSeek-V4-Flash-0731" \
    --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": "deepseek-ai/DeepSeek-V4-Flash-0731",
		"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 "deepseek-ai/DeepSeek-V4-Flash-0731" \
        --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": "deepseek-ai/DeepSeek-V4-Flash-0731",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use deepseek-ai/DeepSeek-V4-Flash-0731 with Docker Model Runner:
```
docker model run hf.co/deepseek-ai/DeepSeek-V4-Flash-0731
```

Resources

View closed (4)

Upload 2 files

#48 opened 6 minutes ago by

Serving recipe: full 3x1M context on 2x NVIDIA DGX Spark (GB10), 3.39M-token KV pool, 653 Tok/s Peak Decode (list content @ c=16)

#47 opened about 8 hours ago by

PSA If you have terrible looping issues or random character generations try update your CUDA to 13.3

#46 opened about 21 hours ago by

This model has many tool_call errors.

#45 opened 1 day ago by

Here's how to run on SM120 / RTX 6000 Pro's (two of them) - not a troll/rickroll, I promise

#44 opened 2 days ago by

Please add recipes and support for Blackwell (sm120) gpus

#43 opened 3 days ago by

Update README.md

#42 opened 3 days ago by

If the actual form of v4-Flash is already strong as such F**K, then what's final version of v4-Pro and WHEN it will be released?

#41 opened 3 days ago by

# 0731 KV cache per-token is ~8× larger than preview (56 vs 7.2 bytes/token) - is this expected?

#40 opened 3 days ago by

dfsafdsafxcvzxv

Reasoning loops

#39 opened 3 days ago by

Add community evaluation results for DEEP-SWE

#38 opened 4 days ago by

Tool-calling instruction is not consistently followed

#37 opened 4 days ago by

[Ktransformers official] add Ktransformers cookbook to model card for consumer hardware

#36 opened 4 days ago by

<3 to deepseek team....

#34 opened 4 days ago by

VISION VISION VISION

#33 opened 5 days ago by

Running 0731 on a single DGX Spark (GB10, 128GB) - working vLLM config, 256K context

#32 opened 5 days ago by

Wen yah

#31 opened 5 days ago by

Update README.md

#30 opened 5 days ago by

Model returned random math problem when given nothing

#29 opened 5 days ago by

[HF Staff] This tag is required to enable inference providers, please merge!

#28 opened 6 days ago by

ds4

#27 opened 6 days ago by

Add missing chat_template.jinja

#26 opened 6 days ago by

梁圣牛逼; deepseek is supercalifragilisticexpialidocious

#25 opened 6 days ago by

震惊瘫坐，仿佛看到了原子弹爆炸！

#24 opened 6 days ago by

GigaSeek

#23 opened 6 days ago by

How can it be served with 4 x RTX Pro6000?

#22 opened 6 days ago by

Preserve reasoning

#21 opened 6 days ago by

Publish a blog or paper on how the improvements were achieved.

#18 opened 7 days ago by

DGX Spark deploy

#17 opened 7 days ago by

This should be named V4.1?

#16 opened 7 days ago by

Are you saying I can deploy a top-tier T1 model with just a few thousand dollars' worth of Mac Studio?

#15 opened 7 days ago by

你的意思是我只需要一个几万人民币的mac studio就可以部署一个T1顶级模型吗？

#14 opened 7 days ago by

multimodal

#13 opened 7 days ago by

梁圣牛逼！！！

#12 opened 7 days ago by

梁圣牛批

#11 opened 7 days ago by

GOAT

#10 opened 7 days ago by

梁圣牛批，国模雄起！

#8 opened 7 days ago by

You rock guys!

#7 opened 7 days ago by

kyrylo-gorbachov

Is dspark already included in these weights?

#6 opened 7 days ago by

Small Model

#5 opened 7 days ago by

国产之光!

#4 opened 7 days ago by

梁圣牛批

#3 opened 7 days ago by

Will you also publish the base model?

#2 opened 7 days ago by

Going completely mad

#1 opened 7 days ago by