deepseek-ai
/

DeepSeek-V4.1-Flash

Image-Text-to-Text

text-generation

8-bit precision

Model card Files Files and versions

Instructions to use deepseek-ai/DeepSeek-V4.1-Flash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use deepseek-ai/DeepSeek-V4.1-Flash with Transformers:

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

pipe = pipeline("image-text-to-text", model="deepseek-ai/DeepSeek-V4.1-Flash")

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-V4.1-Flash", device_map="auto")

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use deepseek-ai/DeepSeek-V4.1-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-V4.1-Flash"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "deepseek-ai/DeepSeek-V4.1-Flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/deepseek-ai/DeepSeek-V4.1-Flash

How to use deepseek-ai/DeepSeek-V4.1-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-V4.1-Flash" \
    --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": "deepseek-ai/DeepSeek-V4.1-Flash",
		"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 "deepseek-ai/DeepSeek-V4.1-Flash" \
        --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": "deepseek-ai/DeepSeek-V4.1-Flash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use deepseek-ai/DeepSeek-V4.1-Flash with Docker Model Runner:
```
docker model run hf.co/deepseek-ai/DeepSeek-V4.1-Flash
```

Resources

View closed (6)

Create app.py

#60 opened about 3 hours ago by

Formal verification of Engram/Conditional Memory — safety review request

#59 opened 3 days ago by

Lowest-latency hosted provider for DeepSeek V4.1 Flash from Northern Virginia (US East)?

#57 opened 4 days ago by

DeepSeek V4.1 Flash VRAM requirements check here: https://huggingface.co/spaces/yash-711/deepseek-v41-flash-fit

#56 opened 4 days ago by

upload 100%

#55 opened 4 days ago by deleted

Upload 71 files

#54 opened 4 days ago by deleted

Update encoding/test_encoding.py

#53 opened 4 days ago by deleted

Streaming this from SSD: check whether your engine reads all 384 experts per layer

#52 opened 4 days ago by

konstantinnikol

Will base model be released?

#51 opened 5 days ago by

Question about benchmark score (deepswe v1.1)

#50 opened 5 days ago by

🚩 Report: Spam

#49 opened 5 days ago by

tester9632587411

Update README.md

#48 opened 5 days ago by

DeepSeek V4.1 Flash manages agent memory 10× less efficiently than V4 Flash (Hermes, MarathonMemBench)

#46 opened 5 days ago by

Add ExtractBench evaluation results

#45 opened 6 days ago by

boyang-runllama

Add ParseBench evaluation results

#44 opened 6 days ago by

boyang-runllama

Successful DeepSeek-V4.1-Flash deployment on 4× A100 80GB using shi3z/deepseekv4.1-A100-custom — memory usage, load time and comparison with vLLM

#43 opened 6 days ago by

marcinmisiaszek

Native Transformers text-backbone support proposal (#48721)

#42 opened 7 days ago by

Independent CPU decode/prefill regression evidence for the index_k fix in PR #12

#41 opened 7 days ago by

Update config.json

#40 opened 8 days ago by

V4.1-Flash: 33–50 agent turns for a one-field config edit (verification with no stopping criterion)

#39 opened 8 days ago by

Reading DeepSeek V4.1 Flash's config.json: Where Sparse Attention Lives

#38 opened 8 days ago by

Runs on one RTX 5090 (31.8 GiB) + 125.7 GiB RAM via a llama.cpp fork: GGUF, report and numbers

#37 opened 9 days ago by

"Le Cerveau dans un Bocal de Morve: DeepSeek-V4.1-Flash or the Art of Selling a 4B Invalid as a Frontier Thinker" 🧠💧🧪

#36 opened 9 days ago by

Tested on coding and agentic work - not that impressive

#35 opened 9 days ago by

Nexeo

#34 opened 9 days ago by

Instantly parse Hugging Face & ModelScope safetensors metadata without downloading weights. View tensor shapes, dtypes, and run side-by-side model diffs.

#33 opened 9 days ago by

smaller model with engram?

#32 opened 9 days ago by

More long-context evaluation results?

#31 opened 10 days ago by

Hey DeepSeek, could you avoid using such confusing model IDs on the API platform?

#30 opened 10 days ago by

Add community evaluation results

#29 opened 10 days ago by

Running on 4x RTX PRO 6000 with NVMe offload for ngram

#28 opened 10 days ago by

Love to see the Harness Benchmark!!! TY!!!

#26 opened 10 days ago by

哇、DeepSeek！

#25 opened 10 days ago by

no way w deepseek

#24 opened 10 days ago by

很强，参数量比上个版本翻倍，最强的flash模型，unsloth 早点出量化版本，赞美这些开源大模型

#22 opened 10 days ago by

感谢Deepseek

#21 opened 10 days ago by

<a href=https://evil.com>hello</a>

#20 opened 10 days ago by

tester9632587411

前排留名

#19 opened 10 days ago by

OpenAI and Claude don't make me download half a terabyte of weight just to ask a question smh

#18 opened 10 days ago by

Has the model's alignment with human ethics been strengthened compared to the previous generation?

#17 opened 10 days ago by

Bro....500多B的Flash，8卡H200已经上不了桌了吗[cry]

#16 opened 10 days ago by

Back to "attention is all you need"

#15 opened 10 days ago by

update README to add vLLM inference

#13 opened 10 days ago by

Restore each indexer's K cache on incomplete compression steps

#12 opened 10 days ago by

DeepSeek V4.1 Flash Lite

#11 opened 10 days ago by

为啥简单问题也强行输出小作文

#9 opened 10 days ago by

Thanks for open-sourcing，感谢开源 DeepSeek V4.1 Flash

#8 opened 10 days ago by

Very impressive!

#7 opened 10 days ago by

Update README.md

#6 opened 10 days ago by

出来溜达一圈，等待社区反馈

#5 opened 10 days ago by