A simple, unified multimodal models training engine. Lean, flexible, and built for hacking at scale.

DDT: Decoupled Diffusion Transformer

Official implementation of "UniLiP: Adapting CLIP for Unified Multimodal Understanding, Generation and Editing"

molly, an LLM designed to understand multi-omics data.

Implementation of "Hyperspherical Latents Improve Continuous-Token Autoregressive"

Pytorch implemention of UniFlow

Official PyTorch Implementation of "Diffusion Transformers with Representation Autoencoders"

Benchmarking Knowledge Transfer in Lifelong Robot Learning

ULMEvalKit: One-Stop Eval ToolKit for Image Generation

“FlowAR: Scale-wise Autoregressive Image Generation Meets Flow Matching” FlowAR employs a simplest scale design and is compatible with any VAE.

Code release for Ming-UniVision: Joint Image Understanding and Geneation with a Continuous Unified Tokenizer

[ICCV 2025] LIRA

HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation

WorldVLA: Towards Autoregressive Action World Model

Official repo for NeurIPS 2025 poster: Unveiling the Spatial-temporal Effective Receptive Fields of Spiking Neural Networks

[NeurIPS-24] This is the official implementation of the paper "DeepStack: Deeply Stacking Visual Tokens is Surprisingly Simple and Effective for LMMs".

[NeurIPS 2025] The official implementation of paper "Safe-Sora: Safe Text-to-Video Generation via Graphical Watermarking"

[CVPR2025 Highlight] PAR: Parallelized Autoregressive Visual Generation. https://yuqingwang1029.github.io/PAR-project

Sequence Parallelism for Long Training

Meta Lingua: a lean, efficient, and easy-to-hack codebase to research LLMs.

Unified Long Training Codebase

Large World Model -- Modeling Text and Video with Millions Context

Fully Open Framework for Democratized Multimodal Training

Multi-Level Triton Runner supporting Python, IR, PTX, and cubin.

[Fully open] [Encoder-free MLLM] Vision as LoRA

OneCAT: Decoder-Only Auto-Regressive Model for Unified Understanding and Generation

Code for our paper "Next Visual Granularity Generation".

[ICCV2025]Code Release of Harmonizing Visual Representations for Unified Multimodal Understanding and Generation

Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models