deepseek-ai
/

DeepSeek-V3.2

Text Generation

Model card Files Files and versions

Instructions to use deepseek-ai/DeepSeek-V3.2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use deepseek-ai/DeepSeek-V3.2 with Transformers:

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

pipe = pipeline("text-generation", model="deepseek-ai/DeepSeek-V3.2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("deepseek-ai/DeepSeek-V3.2")
model = AutoModelForCausalLM.from_pretrained("deepseek-ai/DeepSeek-V3.2")

Inference
HuggingChat
Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use deepseek-ai/DeepSeek-V3.2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "deepseek-ai/DeepSeek-V3.2"
# 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-V3.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/deepseek-ai/DeepSeek-V3.2

How to use deepseek-ai/DeepSeek-V3.2 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-V3.2" \
    --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-V3.2",
		"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-V3.2" \
        --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-V3.2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use deepseek-ai/DeepSeek-V3.2 with Docker Model Runner:
```
docker model run hf.co/deepseek-ai/DeepSeek-V3.2
```

Resources

View closed (5)

Listed on OpenModelMap

#68 opened about 4 hours ago by

OMS Score -- OMS 79.8 (B) -- DeepSeek V3.2 | 5-dimension independent scoring

#67 opened 5 days ago by

OMS Score -- OMS 79.8 (B) -- DeepSeek V3.2 | 5-dimension independent scoring

#66 opened 5 days ago by

OMS Score -- OMS 79.8 (B) -- DeepSeek V3.2 | 5-dimension independent scoring

#65 opened 5 days ago by

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

#64 opened 13 days ago by

Add WildClawBench evaluation result

#62 opened 29 days ago by

Add Open Agent Leaderboard evaluation results

#61 opened about 1 month ago by

Add Claw-Eval evaluation results

#60 opened about 2 months ago by

Create .eval_results/apex-swe.yaml

#59 opened about 2 months ago by

Create .eval_results/apex-agents.yaml

#58 opened about 2 months ago by

Add YC-Bench benchmark result (avg $125,263)

#57 opened 3 months ago by

Create test

#54 opened 3 months ago by

How to make this model uncensored

#53 opened 3 months ago by

Add SWE-Bench Pro evaluation results

#51 opened 4 months ago by

Add GPQA evaluation results

#50 opened 4 months ago by

Add MMLU Pro evaluation results

#49 opened 4 months ago by

Add HLE evaluation results

#48 opened 4 months ago by

Add SWE-Bench Verified evaluation results

#47 opened 4 months ago by

Add MathArena evaluation result for hmmt/hmmt_feb_2026

#46 opened 4 months ago by

JasperDekoninck

Add Terminal-Bench evaluation result (39.6%)

#45 opened 4 months ago by

Add MathArena evaluation result for aime/aime_2026

#44 opened 4 months ago by

JasperDekoninck

Question about format of building search tool prompt

#43 opened 4 months ago by

Add EvasionBench evaluation results

#42 opened 4 months ago by

Request: DOI

#41 opened 5 months ago by

为什么v3.2中没有modeling_deepseek.py文件？

#40 opened 5 months ago by

Upload internal.py

#39 opened 5 months ago by

Ananthusajeev190

Create depseek

#38 opened 5 months ago by

Reasoning and agentic AI deployment

#37 opened 5 months ago by

Update README.md

#36 opened 6 months ago by

Running the model with a dense attention

#35 opened 6 months ago by

Error accessing the model

#34 opened 6 months ago by

DSA Question

#33 opened 6 months ago by

Function calling ability not stable after mutli-rounds conversations

#32 opened 6 months ago by

Default system prompt can hinder thinking-mode performance

#31 opened 6 months ago by

In reasoning mode, the chain of thought (CoT) tends to leak the user's instructions.

#30 opened 6 months ago by

Why doesn't the model generate ｜DSML｜special tokens?

#29 opened 6 months ago by

A light version of DeepSeek around 100 gigabytes? Also DeepSeek Coder Lite?

#28 opened 6 months ago by

May we please get a light-weight Deepseek model?

#27 opened 6 months ago by deleted

Upload bCNx4JhR_400x400.jpg

#26 opened 6 months ago by

Consistent syntax error writing C++ code

#25 opened 7 months ago by

Should the thinking mode in DS v3.2 use the default system prompt?

#24 opened 7 months ago by

Inconsistent <think> and </think> tags

#23 opened 7 months ago by

NVFP4 Checkpoint?

#22 opened 7 months ago by

Add Artificial Analysis evaluations for deepseek-v3-2

#21 opened 7 months ago by

mackenzietechdocs

which tool-call-parser we should use for deepseek-ai/DeepSeek-V3.2

#19 opened 7 months ago by

Got ModuleNotFoundError: No module named 'torch' when running `pip install fast_hadamard_transform`

#18 opened 7 months ago by

main

#17 opened 7 months ago by

model type `deepseek_v32` not found

#16 opened 7 months ago by

Humble request for a stable vLLM/SGLang deployment setup for DeepSeek-V3.2

#15 opened 7 months ago by

Finally.

#14 opened 7 months ago by