deepseek-ai
/

DeepSeek-V4-Flash

Text Generation

8-bit precision

Model card Files Files and versions

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

Libraries

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

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

pipe = pipeline("text-generation", model="deepseek-ai/DeepSeek-V4-Flash")
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")
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-V4-Flash", device_map="auto")

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

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

Use Docker

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

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

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

Resources

View closed (4)

Delete inference/README.md

#50 opened 6 days ago by

Update README.md

#49 opened 6 days ago by

什么时候满血版v4，梁圣！

#48 opened 8 days ago by

fastllm 启动deepseek-v4-flash 回答问题乱码

#47 opened 16 days ago by

Run DeepSeek-V4-Flash locally! 🐳

#46 opened 23 days ago by

Add SkillsBench v1.1 evaluation result

#45 opened 27 days ago by

Listed on OpenModelMap

#44 opened about 1 month ago by

Listed on OpenModelMap

#43 opened about 1 month ago by

Since the DSA indexer selects the top‑k tokens with respect to k, is it really necessary to multiply all tokens by a softmax_scale scalar?

#42 opened about 1 month ago by

Deepseek-v4-flash image support

#41 opened about 1 month ago by

DeepSeek-V4-Flash fine-tuning: can a merged model keep the same FP4+FP8 format as the base model, or is BF16 the only supported merge output?

#40 opened about 2 months ago by

TemporalMesh Transformer: 29.4 PPL at 48% compute — beats Mamba, new open-source architecture

#38 opened about 2 months ago by

I'm confused (v4 Flash vs v3.2)

#37 opened about 2 months ago by

Number of layers in inference mock code

#36 opened about 2 months ago by

🚀 ms-swift Provides DeepSeek-V4 Fine-tuning Practice

#35 opened 2 months ago by

DeepSeek Training Support

#34 opened 2 months ago by

官方是否有兴趣帮忙改进完善一下Apple MLX框架的适配？

#33 opened 2 months ago by

DeepSeek-V4帮我破了《真探3》的案子

#32 opened 3 months ago by

Add Claw-Eval evaluation results

#31 opened 3 months ago by

use vLLM with 2 DGX Sparks , response broken

#29 opened 3 months ago by

Running models with vLLM on the RTX Pro 6000 - SM120

#28 opened 3 months ago by

model anwser with <ds_safety>

#27 opened 3 months ago by

Claude 4.7 Reasoning into DeepSeek-V4-Flash for Open-Source Edge Intelligence

#26 opened 3 months ago by

muneebkhanniazi

How to run deepseek on Ada GPUs？Mine is L20.

#25 opened 3 months ago by

How can I use this with llama.cpp?

#24 opened 3 months ago by

Questions on MoE Hash Routing

#22 opened 3 months ago by

120B model?

#21 opened 3 months ago by

Run DeepSeek-V4-Flash on more hardware: FP8/BF16 adapted versions for 8 AI chips (ready to download)

#20 opened 3 months ago by

[适配请求] 建议 API 兼容 OpenAI 新规范中的 developer 角色（It is recommended that the API be compatible with the developer role in the new OpenAI specification

#19 opened 3 months ago by

Update README.md to add SGLang deployment option

#18 opened 3 months ago by

Is 158B or 284b params ?

#17 opened 3 months ago by

Add chat template

#16 opened 3 months ago by

Unable to run on 2x RTX Pro 6000 (DEEP_GEMM problem)

#15 opened 3 months ago by

bsarpel

#14 opened 3 months ago by

Local Installation Video and Testing - Step by Step

#13 opened 3 months ago by

Too big to run locally.

#12 opened 3 months ago by

Add community evaluation results for GPQA, MMLU-PRO, SWE-BENCH_VERIFIED, HLE, TERMINAL-BENCH-2.0

#11 opened 3 months ago by

May I ask if there is a deployment document?

#10 opened 3 months ago by

Will there be a smaller model like Qwen3.5 122 or Nemotron 3 super

#9 opened 3 months ago by

终于来了~

#7 opened 3 months ago by

量化相关

#6 opened 3 months ago by

起立，立正！

#5 opened 3 months ago by

前排前排！合影

#4 opened 3 months ago by

Sick

#3 opened 3 months ago by

Green-eyedDevil

support inference by sglang or vllm ?

#2 opened 3 months ago by

Here comes the big one

#1 opened 3 months ago by