README.md

---
license: mit
---
# DeepSeek-V3.2-Retro

This repository hosts the model weights for **DeepSeek-V3.2-Retro**. For instructions and details, please refer to the [GitHub](https://github.com/zhejianglab/DeepSeek-V3.2-Retro).

## 1. Introduction
[DeepSeek-V3.2](https://huggingface.co/deepseek-ai/DeepSeek-V3.2) 
introduces the DeepSeek Sparse Attention (DSA) architecture, representing a significant architectural evolution over [DeepSeek-V3](https://huggingface.co/deepseek-ai/DeepSeek-V3) and [DeepSeek-V3.1](https://huggingface.co/deepseek-ai/DeepSeek-V3.1). However, as of now, an official open-source implementation compatible with Ampere-series GPUs has not been released.

To address this gap, we introduce **DeepSeek-V3.2-Retro**, targeting the following user groups:

- Ampere GPU users who do not have access to Hopper or Blackwell architectures.
- Users of general-purpose GPU platforms where DSA is not yet supported.

Key features of **DeepSeek-V3.2-Retro** include:

- Removal of the DSA modules from the original V3.2 architecture.
- Conversion of model parameters and computation to the BF16 data format.
- Broad Compatibility: runs on any hardware platform that supports the V3 architecture.
- Validated Performance: achieves performance on multiple benchmarks that is close to the [officially reported results](https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/main/assets/paper.pdf).

## 2. Performance Evaluation
As our primary target scenario is reasoning-oriented usage, we report accuracy results on several representative benchmarks after enabling the thinking feature. All evaluation metrics are taken from the corresponding official technical reports for consistency.

<div align="center">

| Benchmark | [DeepSeek-V3.2-Retro](https://github.com/zhejianglab/DeepSeek-V3.2-Retro) | [DeepSeek-V3.2-Thinking](https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/main/assets/paper.pdf) |
| :---: | :---: | :---: |
| MMLU-Pro | 86.4 | 85.0 |
| GPQA Diamond | 82.12 | 82.4 |
| AIME 2025 | 93.67 | 93.1 |
| LiveCodeBench | 80.72 | 83.3 |

</div>

In addition, we evaluate inference efficiency. Using SGLang v0.5.6 under identical settings, we observe that the throughput of DeepSeek-V3.2-Retro is on par with DeepSeek-V3.1. Output throughput is reported in tokens/s.

<div align="center">

| Model | Output Throughput (qps=512, input=1k, output=10k) |
| :---: | :---: | 
| [DeepSeek-V3.2-Retro](https://github.com/zhejianglab/DeepSeek-V3.2-Retro) | 2510.27 | 
| [DeepSeek-V3.1](https://huggingface.co/deepseek-ai/DeepSeek-V3.1) | 2515.34 | 

</div>

These results indicate that removing the DSA structure and reverting to a V3-compatible architecture does not introduce noticeable performance regression in either reasoning accuracy or inference throughput on Ampere-class hardware.

## 3. Model Download
DeepSeek-V3.2-Retro model is available for download from [Hugging Face](https://huggingface.co/ZhejiangLab/DeepSeek-V3.2-Retro) and [ModelScope](https://modelscope.cn/models/zhejianglab/DeepSeek-V3.2-Retro). Please ensure that you have at least 1.5 TB of available disk space before downloading the model.

<div align="center">

| **Model** | **Total Params** | **Hugging Face** | **ModelScope** | 
|:---------:|:----------------:|:----------------:|:--------------:|
| DeepSeek-V3.2-Retro | 684 B | [🤗 Hugging Face](https://huggingface.co/ZhejiangLab/DeepSeek-V3.2-Retro) |[🤖 ModelScope](https://modelscope.cn/models/zhejianglab/DeepSeek-V3.2-Retro) |

</div>

## 4. Quickstart

We strongly recommend using SGLang for efficient inference of the DeepSeek series models. We provide example configurations for SGLang serving on four A100*8 nodes.

### SGLang

#### Using Docker (Recommended)

```docker
# Pull latest image on four nodes and ensure RDMA network connectivity between the 4 nodes.
# https://hub.docker.com/r/lmsysorg/sglang/tags
docker pull lmsysorg/sglang:latest
```

#### Launch Command

```python
# For high QPS scenarios, add --enable-dp-attention and --ep-size arguments to boost throughput, and use mtp to boost decoding speed.
# node 1
python3 -m sglang.launch_server --model-path /path/to/DeepSeek-V3.2-Retro --tp 32 --dist-init-addr 10.0.0.1:5000 --nnodes 4 --node-rank 0 --trust-remote-code --host 0.0.0.0 --port 30000 --speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 --enable-dp-attention --dp 8 --ep-size 32 --enable-dp-lm-head

# node 2
python3 -m sglang.launch_server --model-path /path/to/DeepSeek-V3.2-Retro --tp 32 --dist-init-addr 10.0.0.1:5000 --nnodes 4 --node-rank 1 --trust-remote-code --speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 --enable-dp-attention --dp 8 --ep-size 32 --enable-dp-lm-head

# node 3
python3 -m sglang.launch_server --model-path /path/to/DeepSeek-V3.2-Retro --tp 32 --dist-init-addr 10.0.0.1:5000 --nnodes 4 --node-rank 2 --trust-remote-code --speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 --enable-dp-attention --dp 8 --ep-size 32 --enable-dp-lm-head

# node 4
python3 -m sglang.launch_server --model-path /path/to/DeepSeek-V3.2-Retro --tp 32 --dist-init-addr 10.0.0.1:5000 --nnodes 4 --node-rank 3 --trust-remote-code --speculative-algorithm NEXTN --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 --enable-dp-attention --dp 8 --ep-size 32 --enable-dp-lm-head
```

## 5. License
This repository and the model weights are licensed under the MIT License, following the license of DeepSeek-V3.2. In addition, if you use DeepSeek-V3.2, you shall also comply with the terms and conditions of DeepSeek-V3.2.

## 6. Contact
If you have any questions, please raise an [issue](https://github.com/zhejianglab/DeepSeek-V3.2-Retro/issues) or contact us at opensource@zhejianglab.org.