OmniVinci is an omni-modal LLM for joint understanding of vision, audio, and language.

🚀🚀 「大模型」2小时完全从0训练26M的小参数GPT！🌏 Train a 26M-parameter GPT from scratch in just 2h!

Native Multimodal Models are World Learners

UniVid: The Open-Source Unified Video Model

my commonly-used tools

Official repository of paper "LOVE-R1: Advancing Long Video Understanding with Adaptive Zoom-in Mechanism via Multi-Step Reasoning"

Implementation of a single layer of the MMDiT, proposed in Stable Diffusion 3, in Pytorch

Official Implementation of "UniFlow: A Unified Pixel Flow Tokenizer for Visual Understanding and Generation"

[NeurIPS 2025 D&B🔥] ImgEdit: A Unified Image Editing Dataset and Benchmark

An official implementation of "CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning"

code for "Diffusion Forcing: Next-token Prediction Meets Full-Sequence Diffusion"

Inference script for Oasis 500M

A benchmark for evaluating vision-centric, complex video reasoning.

Open-source unified multimodal model

Implementation of Denoising Diffusion Probabilistic Model in Pytorch

Official PyTorch Implementation of "Scalable Diffusion Models with Transformers"

Structured Video Comprehension of Real-World Shorts

[ICCV 2025] LVBench: An Extreme Long Video Understanding Benchmark

Official repo and evaluation implementation of VSI-Bench

【干货】史上最全的PyTorch学习资源汇总

The Next Step Forward in Multimodal LLM Alignment

Official code for MotionBench (CVPR 2025)

Ultra-high-performance, secure, all-in-one acceleration engine for developer resources

个人构建MoE大模型：从预训练到DPO的完整实践

Tracking the latest and greatest research papers on video generation.

Long-RL: Scaling RL to Long Sequences (NeurIPS 2025)

Agent-R1: Training Powerful LLM Agents with End-to-End Reinforcement Learning