deepseek-ai
/

DeepSeek-V3.1

Text Generation

text-generation-inference

Model card Files Files and versions

Instructions to use deepseek-ai/DeepSeek-V3.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use deepseek-ai/DeepSeek-V3.1 with Transformers:

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

pipe = pipeline("text-generation", model="deepseek-ai/DeepSeek-V3.1", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V3.1", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-V3.1", trust_remote_code=True, device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use deepseek-ai/DeepSeek-V3.1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "deepseek-ai/DeepSeek-V3.1"
# 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-V3.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/deepseek-ai/DeepSeek-V3.1

How to use deepseek-ai/DeepSeek-V3.1 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-V3.1" \
    --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-V3.1",
		"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-V3.1" \
        --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-V3.1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use deepseek-ai/DeepSeek-V3.1 with Docker Model Runner:
```
docker model run hf.co/deepseek-ai/DeepSeek-V3.1
```

Resources

View closed (9)

Listed on OpenModelMap

#53 opened about 1 month ago by

OMS Score -- OMS 78.8 (B) -- DeepSeek V3.1 | 5-dimension independent scoring

#52 opened about 2 months ago by

Add EvalEval community eval results

#51 opened 2 months ago by

How is routed_scaling_factor = 2.5 chosen?

#50 opened 2 months ago by

Fix chat_template crash when assistant message omits the `content` key

#49 opened 3 months ago by

Distill the model into 4 variants: 600M, 1.5B, 4B and 7B

#48 opened 10 months ago by

模型在复杂工具调用出现输出乱码的

#47 opened 11 months ago by

Quantum Ai

#45 opened 11 months ago by

调用不存在的 tool

#43 opened 11 months ago by

中文 markdown 表格容易出错

#42 opened 11 months ago by

OUTPUT_PARSING_FAILURE

#41 opened 11 months ago by

RuntimeError: No DeepseekV2MoE layer found in model.layers.

#40 opened 11 months ago by

Script to download deepseek tensor files

#39 opened 11 months ago by

Criticism, please read

#38 opened 11 months ago by

Solve this Isuue Please

#37 opened 11 months ago by

多轮tool call能力不稳定

#35 opened 11 months ago by

Have you guys noticed this "极""極""extreme"problem?

#34 opened 11 months ago by

为什么模型卡片里的weight_scale_inv即使用fp32存储，也还是存在尾数？

#33 opened 11 months ago by

Feat: Add toolcall description

#32 opened 11 months ago by

Update README.md

#29 opened 12 months ago by

[W4A8 FP8 Quantization] Release of DeepSeek-V3.1 with SGLang Support – Near-Lossless & 1.56x Speed Boost!

#28 opened 12 months ago by

tool call for reasoning mode

#27 opened 12 months ago by

V3.1 seems to be pretty bad at everything except coding and mathematics. V3.1 看起来除了编程和数学之外，其他方面都很差。

#26 opened 12 months ago by

DeepSeek-V3.1全方位最新实测出炉（300+维度），欢迎进群交流讨论~

#25 opened 12 months ago by

有人尝试过本地通过vllm部署并连接到claude code吗？

#24 opened 12 months ago by

why all the scale values of attn out_proj have the none-zero mantissa 0b111010101010010101001111111101 ?

#22 opened 12 months ago by

Update system prompt to include tools

#21 opened 12 months ago by

recommended temp?

#19 opened 12 months ago by

Tool calling usage examples

#18 opened 12 months ago by

Context length: is it 128K (as mentioned in the model card) or 160K (as specified in config.json)?

#17 opened 12 months ago by

不知道还有没有蒸馏模型

#13 opened 12 months ago by

no score on swe under thinking mode

#11 opened 12 months ago by

This model’s censorship is insane

#10 opened 12 months ago by

请问simpleQA结果是否为笔误？

#9 opened 12 months ago by

梁文疯垃圾模型

#7 opened 12 months ago by

search是用think模式吗

#6 opened 12 months ago by

Congratulations to DeepSeek, this version seems powerful for coding and agent development

#5 opened 12 months ago by

火速下载

#4 opened 12 months ago by

Any plan to release the post-training recipe?

#3 opened 12 months ago by

Come on, third party bros! Deploy it!

#2 opened 12 months ago by

火前留名

#1 opened 12 months ago by