[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-72614":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":19,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":27,"trendingCount":16,"starSnapshotCount":16,"syncStatus":18,"lastSyncTime":28,"discoverSource":29},72614,"VideoSys","NUS-HPC-AI-Lab\u002FVideoSys","NUS-HPC-AI-Lab","VideoSys: An easy and efficient system for video generation","",null,"Python",2025,130,23,22,0,1,2,3,61.65,"Apache License 2.0",false,"master",[],"2026-06-12 04:01:06","\u003Cp align=\"center\">\n\u003Cimg width=\"55%\" alt=\"VideoSys\" src=\".\u002Fassets\u002Ffigures\u002Flogo.png?raw=true\">\n\u003C\u002Fp>\n\u003Ch3 align=\"center\">\nAn easy and efficient system for video generation\n\u003C\u002Fh3>\n\u003Cp align=\"center\">| \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys?tab=readme-ov-file#installation\">Quick Start\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys?tab=readme-ov-file#usage\">Supported Models\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys?tab=readme-ov-file#acceleration-techniques\">Accelerations\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fdiscord.gg\u002FWhPmYm9FeG\">Discord\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Foahzxl.notion.site\u002FVideoSys-News-42391db7e0a44f96a1f0c341450ae472?pvs=4\">Media\u003C\u002Fa> | \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FVideoSys\">HuggingFace Space\u003C\u002Fa> |\n\u003C\u002Fp>\n\n### Latest News 🔥\n- [2024\u002F11] 🔥 \u003Cb>Propose Data-Centric Parallel (DCP) [[blog](https:\u002F\u002Foahzxl.github.io\u002FDCP\u002F)][[doc](.\u002Fdocs\u002Fdcp.md)], a simple and efficient method for variable sequences (\u003Ci>e.g., \u003C\u002Fi> videos) training\u003C\u002Fb>.\n- [2024\u002F09] Support [CogVideoX](https:\u002F\u002Fgithub.com\u002FTHUDM\u002FCogVideo), [Vchitect-2.0](https:\u002F\u002Fgithub.com\u002FVchitect\u002FVchitect-2.0) and [Open-Sora-Plan v1.2.0](https:\u002F\u002Fgithub.com\u002FPKU-YuanGroup\u002FOpen-Sora-Plan).\n- [2024\u002F08] 🔥 Evole from [OpenDiT](https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys\u002Ftree\u002Fv1.0.0) to \u003Cb>VideoSys: An easy and efficient system for video generation\u003C\u002Fb>.\n- [2024\u002F08] 🔥 Release PAB paper: \u003Cb>[Real-Time Video Generation with Pyramid Attention Broadcast](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.12588)\u003C\u002Fb>.\n- [2024\u002F06] 🔥 Propose Pyramid Attention Broadcast (PAB) [[paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.12588)][[blog](https:\u002F\u002Foahzxl.github.io\u002FPAB\u002F)][[doc](.\u002Fdocs\u002Fpab.md)], the first approach to achieve \u003Cb>real-time\u003C\u002Fb> DiT-based video generation, delivering \u003Cb>negligible quality loss\u003C\u002Fb> without \u003Cb>requiring any training\u003C\u002Fb>.\n- [2024\u002F06] Support [Open-Sora-Plan](https:\u002F\u002Fgithub.com\u002FPKU-YuanGroup\u002FOpen-Sora-Plan) and [Latte](https:\u002F\u002Fgithub.com\u002FVchitect\u002FLatte).\n- [2024\u002F03] 🔥 Propose Dynamic Sequence Parallel (DSP)[[paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.10266)][[doc](.\u002Fdocs\u002Fdsp.md)], achieves **3x** speed for training and **2x** speed for inference in Open-Sora compared with sota sequence parallelism.\n- [2024\u002F03] Support [Open-Sora](https:\u002F\u002Fgithub.com\u002Fhpcaitech\u002FOpen-Sora).\n- [2024\u002F02] 🎉 Release [OpenDiT](https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys\u002Ftree\u002Fv1.0.0): An Easy, Fast and Memory-Efficent System for DiT Training and Inference.\n\n# About\n\nVideoSys is an open-source project that provides a user-friendly and high-performance infrastructure for video generation. This comprehensive toolkit will support the entire pipeline from training and inference to serving and compression.\n\nWe are committed to continually integrating cutting-edge open-source video models and techniques. Stay tuned for exciting enhancements and new features on the horizon!\n\n## Installation\n\nPrerequisites:\n\n- Python >= 3.10\n- PyTorch >= 1.13 (We recommend to use a >2.0 version)\n- CUDA >= 11.6\n\nWe strongly recommend using Anaconda to create a new environment (Python >= 3.10) to run our examples:\n\n```shell\nconda create -n videosys python=3.10 -y\nconda activate videosys\n```\n\nInstall VideoSys:\n\n```shell\ngit clone https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys\ncd VideoSys\npip install -e .\n```\n\n\n## Usage\n\nVideoSys supports many diffusion models with our various acceleration techniques, enabling these models to run faster and consume less memory.\n\n\u003Cb>You can find all available models and their supported acceleration techniques in the following table. Click `Code` to see how to use them.\u003C\u002Fb>\n\n\u003Ctable>\n    \u003Ctr>\n        \u003Cth rowspan=\"2\">Model\u003C\u002Fth>\n        \u003Cth rowspan=\"2\">Train\u003C\u002Fth>\n        \u003Cth rowspan=\"2\">Infer\u003C\u002Fth>\n        \u003Cth colspan=\"3\">Acceleration Techniques\u003C\u002Fth>\n    \u003C\u002Ftr>\n    \u003Ctr>\n        \u003Cth>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys?tab=readme-ov-file#dyanmic-sequence-parallelism-dsp-paperdoc\">DSP\u003C\u002Fa>\u003C\u002Fth>\n        \u003Cth>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys?tab=readme-ov-file#pyramid-attention-broadcast-pab-blogdoc\">PAB\u003C\u002Fa>\u003C\u002Fth>\n        \u003Cth>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys?tab=readme-ov-file#pyramid-attention-broadcast-pab-blogdoc\">DCP\u003C\u002Fa>\u003C\u002Fth>\n    \u003C\u002Ftr>\n    \u003Ctr>\n        \u003Ctd>Vchitect [\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FVchitect\u002FVchitect-2.0\">source\u003C\u002Fa>]\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u003Ca href=\".\u002Fexamples\u002Finference\u002Fvchitect\u002Fsample.py\">Code\u003C\u002Fa>\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n        \u003Ctd>CogVideoX [\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FTHUDM\u002FCogVideo\">source\u003C\u002Fa>]\u003C\u002Ftd>\n        \u003Ctd align=\"center\">🟡\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u003Ca href=\".\u002Fexamples\u002Finference\u002Fcogvideox\u002Fsample.py\">Code\u003C\u002Fa>\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">🟡\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n        \u003Ctd>Latte [\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FVchitect\u002FLatte\">source\u003C\u002Fa>]\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u003Ca href=\".\u002Fexamples\u002Finference\u002Flatte\u002Fsample.py\">Code\u003C\u002Fa>\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n        \u003Ctd>Open-Sora-Plan [\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FPKU-YuanGroup\u002FOpen-Sora-Plan\">source\u003C\u002Fa>]\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u003Ca href=\".\u002Fexamples\u002Finference\u002Fopen_sora_plan\u002Fsample.py\">Code\u003C\u002Fa>\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u002F\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n        \u003Ctd>Open-Sora [\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fhpcaitech\u002FOpen-Sora\">source\u003C\u002Fa>]\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u003Ca href=\".\u002Fexamples\u002Ftraining\u002Fopen_sora\">Code\u003C\u002Fa>\u003C\u002Ftd>\n        \u003Ctd align=\"center\">\u003Ca href=\".\u002Fexamples\u002Finference\u002Fopen_sora\u002Fsample.py\">Code\u003C\u002Fa>\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n        \u003Ctd align=\"center\">✅\u003C\u002Ftd>\n    \u003C\u002Ftr>\n\u003C\u002Ftable>\n\nYou can also find easy demo with HuggingFace Space \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FVideoSys\">[link]\u003C\u002Fa> and Gradio \u003Ca href=\".\u002Fgradio\">[link]\u003C\u002Fa>. 🟡 means work in progress.\n\n## Acceleration Techniques\n\n### Data-Centric Parallel (DCP) [[blog](https:\u002F\u002Foahzxl.github.io\u002FDCP\u002F)][[doc](.\u002Fdocs\u002Fdcp.md)]\n\n\u003C!-- ![method](.\u002Fassets\u002Ffigures\u002Fdcp_overview.png) -->\n\u003Cp align=\"center\">\n    \u003Cimg src=\".\u002Fassets\u002Ffigures\u002Fdcp_overview.png\" alt=\"method\" height=\"300\">\n\u003C\u002Fp>\nData-Centric Parallel (DCP) is a simple but effective approach to accelerate distributed training of variable sequences. Unlike previous methods that fix training settings, DCP dyanmically adjusts parallelism and other configs driven by incoming data during runtime, achieving up to 2.1x speedup. As a ease-of-use method, DCP can enpower any video models and parallel methods with minimal code changes.\n\nSee its details [here](.\u002Fdocs\u002Fdcp.md).\n\n----\n\n### Pyramid Attention Broadcast (PAB) [[paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.12588)][[blog](https:\u002F\u002Foahzxl.github.io\u002FPAB\u002F)][[doc](.\u002Fdocs\u002Fpab.md)]\n\n![method](.\u002Fassets\u002Ffigures\u002Fpab_method.png)\n\nPAB is the first approach to achieve \u003Cb>real-time\u003C\u002Fb> DiT-based video generation, delivering \u003Cb>lossless quality\u003C\u002Fb> without \u003Cb>requiring any training\u003C\u002Fb>. By mitigating redundant attention computation, PAB achieves up to 21.6 FPS with 10.6x acceleration, without sacrificing quality across popular DiT-based video generation models including [Open-Sora](https:\u002F\u002Fgithub.com\u002Fhpcaitech\u002FOpen-Sora), [Latte](https:\u002F\u002Fgithub.com\u002FVchitect\u002FLatte) and [Open-Sora-Plan](https:\u002F\u002Fgithub.com\u002FPKU-YuanGroup\u002FOpen-Sora-Plan).\n\nSee its details [here](.\u002Fdocs\u002Fpab.md).\n\n----\n\n### Dyanmic Sequence Parallelism (DSP) [[paper](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.10266)][[doc](.\u002Fdocs\u002Fdsp.md)]\n\n![dsp_overview](.\u002Fassets\u002Ffigures\u002Fdsp_overview.png)\n\nDSP is a novel, elegant and super efficient sequence parallelism for [Open-Sora](https:\u002F\u002Fgithub.com\u002Fhpcaitech\u002FOpen-Sora), [Latte](https:\u002F\u002Fgithub.com\u002FVchitect\u002FLatte) and other multi-dimensional transformer architecture.\n\nIt achieves **3x** speed for training and **2x** speed for inference in Open-Sora compared with sota sequence parallelism ([DeepSpeed Ulysses](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.14509)). For a 10s (80 frames) of 512x512 video, the inference latency of Open-Sora is:\n\n| Method | 1xH800 | 8xH800 (DS Ulysses) | 8xH800 (DSP) |\n| ------ | ------ | ------ | ------ |\n| Latency(s) | 106 | 45 | 22 |\n\nSee its details [here](.\u002Fdocs\u002Fdsp.md).\n\n\n## Contributing\n\nWe welcome and value any contributions and collaborations. Please check out [CONTRIBUTING.md](.\u002FCONTRIBUTING.md) for how to get involved.\n\n## Contributors\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys\u002Fgraphs\u002Fcontributors\">\n  \u003Cimg src=\"https:\u002F\u002Fcontrib.rocks\u002Fimage?repo=NUS-HPC-AI-Lab\u002FVideoSys\"\u002F>\n\u003C\u002Fa>\n\n## Star History\n\n[![Star History Chart](https:\u002F\u002Fapi.star-history.com\u002Fsvg?repos=NUS-HPC-AI-Lab\u002FVideoSys&type=Date)](https:\u002F\u002Fstar-history.com\u002F#NUS-HPC-AI-Lab\u002FVideoSys&Date)\n\n## Citation\n\n```\n@misc{videosys2024,\n  author={VideoSys Team},\n  title={VideoSys: An Easy and Efficient System for Video Generation},\n  year={2024},\n  publisher={GitHub},\n  url = {https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys},\n}\n\n@misc{zhao2024pab,\n  title={Real-Time Video Generation with Pyramid Attention Broadcast},\n  author={Xuanlei Zhao and Xiaolong Jin and Kai Wang and Yang You},\n  year={2024},\n  eprint={2408.12588},\n  archivePrefix={arXiv},\n  primaryClass={cs.CV},\n  url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.12588},\n}\n\n@misc{zhao2024dsp,\n  title={DSP: Dynamic Sequence Parallelism for Multi-Dimensional Transformers},\n  author={Xuanlei Zhao and Shenggan Cheng and Chang Chen and Zangwei Zheng and Ziming Liu and Zheming Yang and Yang You},\n  year={2024},\n  eprint={2403.10266},\n  archivePrefix={arXiv},\n  primaryClass={cs.DC},\n  url={https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.10266},\n}\n\n@misc{zhao2024opendit,\n  author={Xuanlei Zhao, Zhongkai Zhao, Ziming Liu, Haotian Zhou, Qianli Ma, and Yang You},\n  title={OpenDiT: An Easy, Fast and Memory-Efficient System for DiT Training and Inference},\n  year={2024},\n  publisher={GitHub},\n  url={https:\u002F\u002Fgithub.com\u002FNUS-HPC-AI-Lab\u002FVideoSys\u002Ftree\u002Fv1.0.0},\n}\n```\n","VideoSys是一个用于视频生成的简易高效系统。它采用Python语言开发，支持多种先进的视频生成模型如CogVideoX、Vchitect-2.0等，并引入了诸如动态序列并行（DSP）和金字塔注意力广播（PAB）等加速技术，显著提升了训练与推理速度，同时保持高质量输出。此外，该项目还强调数据为中心的并行方法以优化变长序列处理效率。适用于需要快速生成高质量视频内容的各种场景，包括但不限于创意设计、教育材料制作以及娱乐产业中的自动化内容生产。","2026-06-11 03:42:49","high_star"]