---
title: DeepSeek-V3.2-W8A8
canonical_url: "https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V3.2-W8A8"
md_url: "https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V3.2-W8A8.md"
repository: vllm-ascend/DeepSeek-V3.2-W8A8
chinese_name: vllm-ascend/DeepSeek-V3.2-W8A8
last_updated: 2025-12-18
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - deepseek_v32
architectures:
  - DeepseekV32ForCausalLM
parameters: 685.7B
tensor_type:
  - I8
  - F32
  - BF16
  - I32
library_name:
  - safetensors
  - pytorch
frameworks:
  - Pytorch
downloads: 13894
stars: 1
---

# DeepSeek-V3.2-W8A8

> DeepSeek-V3.2-W8A8 - vllm-ascend 在 ModelScope 开源的模型。DeepSeek-V3.2-w8a8 Model

- **Repository**: vllm-ascend/DeepSeek-V3.2-W8A8
- **Tasks**: text-generation
- **Parameters**: 685.7B
- **Downloads**: 13894
- **Stars**: 1
- **Last updated**: 2025-12-18

Source: https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V3.2-W8A8

---

# DeepSeek-V3.2-w8a8 Model

This is a quantized version of DeepSeek-V3.2 with W8A8 quantization.

## Model Details

- **Model Name**: DeepSeek-V3.2-w8a8-QuaRot
- **Model Type**: Quantized Language Model
- **Quantization**: W8A8 (Weight 8-bit, Activation 8-bit)
- **Base Model**: DeepSeek-V3.2

## Usage

This model is converted to ModelScope format for use with ModelScope libraries.

## Files

- `config.json`: Model configuration
- `quant_model_weights-*.safetensors`: Quantized model weights
- `tokenizer.json`: Tokenizer file
