Welcome to VeOmni#
VeOmni is a versatile framework for both single- and multi-modal pre-training and post-training. It empowers users to seamlessly scale models of any modality across various accelerators, offering both flexibility and user-friendliness.
Get Started
Usage
- Arguments API Reference
- Basic Modules
- Multimodal Data Processing
- Usage of dyn_bsz
- Support New Models — Guide and Checklist
- Qwen3-VL MoE Integration Example
- Qwen3-Omni-MoE Integration Example
- Support New DiT Models — Guide and Reference
- Checkpoint Conversion
- Trainer
- Agent Workflow Guide
- HDFS FUSE Patch for DCP Consolidation
- VeOmni Test Suite Overview
Hardware Support
- Get Started with Ascend NPU
- Typical Usage: Qwen3-VL 8B Training on Ascend NPU
- Ascend Environment Variables
- Precision Analysis and Troubleshooting Guide
- Model Optimization - Profiling Collection, Analysis and Optimization Ideas
- Ascend A2 Docker Image Build and Usage Guide
- Ascend A3 Docker Image Build and Usage Guide
- Ascend Docker Overview
- Ascend Docker Supported Tags
- FAQ: Common Issues and Solutions for Ascend NPU
- VeOmni on AMD ROCm
- VeOmni on Cambricon MLU
Examples
- Qwen3 training guide
- Qwen3.5 training guide
- Qwen3 MoE training guide
- Qwen3 VL training guide
- Qwen3 Omni MoE training guide
- Qwen3-Omni training with offline-extracted audio-enabled video
- MiniMax H3 FL2VA Quick Start
- Wan2.1-I2V training guide
- Wan2.1-T2V Training Guide
- LTX-2.3 training guide
- Qwen3 DPO training guide
- Seed-OSS training guide
Key Features
Design
Transformers v5 Updates
Citation#
If you find VeOmni useful for your research and applications, feel free to give us a star ⭐ or cite us using:
@article{ma2025veomni,
title={VeOmni: Scaling Any Modality Model Training with Model-Centric Distributed Recipe Zoo},
author={Ma, Qianli and Zheng, Yaowei and Shi, Zhelun and Zhao, Zhongkai and Jia, Bin and Huang, Ziyue and Lin, Zhiqi and Li, Youjie and Yang, Jiacheng and Peng, Yanghua and others},
journal={arXiv preprint arXiv:2508.02317},
year={2025}
}
About ByteDance Seed Team#
Founded in 2023, ByteDance Seed Team is dedicated to crafting the industry’s most advanced AI foundation models. The team aspires to become a world-class research team and make significant contributions to the advancement of science and society. You can get to know Bytedance Seed better through the following channels👇
