---
title: hunyuanvideo-modelopt-fp8-sglang-transformer
canonical_url: "https://www.modelscope.cn/models/lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer"
md_url: "https://www.modelscope.cn/models/lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer.md"
repository: lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer
last_updated: 2026-04-28
base_model:
  - hunyuanvideo-community/HunyuanVideo
base_model_relation: quantized
parameters: 12.8B
tensor_type:
  - BF16
  - F8_E4M3
  - F32
library_name:
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
downloads: 46
stars: 0
tags:
  - sglang
  - diffusion
  - modelopt
  - fp8
  - transformer
---

# hunyuanvideo-modelopt-fp8-sglang-transformer

> hunyuanvideo-modelopt-fp8-sglang-transformer - lmsys 在 ModelScope 开源的模型。hunyuanvideo-modelopt-fp8-sglang-transformer

lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer 是 ModelScope 魔搭社区上的 12.8B 参数机器学习模型，基于 hunyuanvideo-community/HunyuanVideo 构建。

- **Repository**: lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer
- **Parameters**: 12.8B
- **Base model**: hunyuanvideo-community/HunyuanVideo
- **Tags**: sglang, diffusion, modelopt, fp8, transformer
- **Downloads**: 46
- **Stars**: 0
- **Last updated**: 2026-04-28

Source: https://www.modelscope.cn/models/lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer

---

# hunyuanvideo-modelopt-fp8-sglang-transformer

This repository contains the SGLang-native ModelOpt FP8 transformer override for `hunyuanvideo-community/HunyuanVideo`.

It is intended to be used with SGLang Diffusion through `--transformer-path` while keeping the base model separate:

```bash
sglang generate \
  --model-path hunyuanvideo-community/HunyuanVideo \
  --transformer-path lmsys/hunyuanvideo-modelopt-fp8-sglang-transformer \
  --prompt "A cinematic scene with detailed lighting" \
  --save-output
```

## Contents

The repository is intentionally minimal and contains only:

- `config.json`
- `*.safetensors` weight shard files
- `*.safetensors.index.json` when the checkpoint is sharded

Validation images, benchmark outputs, profiler traces, and conversion scratch artifacts are not stored in this model repository.

## Notes

- Quantization config is stored in `config.json` with `quant_method=modelopt` and `quant_algo=FP8`.
- Use this checkpoint with an SGLang version that includes diffusion ModelOpt support for the corresponding model family.
- The original base model license and usage terms still apply.