Tzu-Heng Lin

Kimi-K3. Open Frontier Intelligence

flashlib. Fast and memory-efficient classical machine learning operators

sparser-faster-llms. Cuda kernels for leveraging LLM sparsity to improve throughput and decrease the memory requirements during inference and training.

autoresearch. AI agents running research on single-GPU nanochat training automatically

MiMo-Code. MiMo Code: Where Models and Agents Co-Evolve

diffusion-pipe. A pipeline parallel training script for diffusion models.

Bernini. Bernini is a unified framework for video generation and editing that combines an MLLM-based semantic planner with a DiT-based renderer.

PiD. PiD: Fast and High-Resolution Latent Decoding with Pixel Diffusion

vggt-omega. [CVPR 2026 Oral] VGGT Omega

seedance-2.0. Comprehensive production pipeline for quad-modal AI filmmaking with Seedance 2.0

modded-nanogpt. NanoGPT (124M) in 90 seconds

pytorch-memory-fix. Two environment variables that fix PyTorch/glibc memory creep on Linux. Zero code changes. Zero performance cost.

pm-skills. PM Skills Marketplace: 100+ agentic skills, commands, and plugins — from discovery to strategy, execution, launch, and growth.

ms-swift. Use PEFT or Full-parameter to CPT/SFT/DPO/GRPO 600+ LLMs (Qwen3.6, DeepSeek-V4, GLM-5.1, InternLM3, Llama4, ...) and 300+ MLLMs (Qwen3-VL, Qwen3-Omni, InternVL3.5, Ovis2.5, GLM4.5v, Gemma4, Llava, Phi4, ...) (AAAI 2025).

qust. HTML

waoowaoo. 首家工业级全流程 AI 影视生产平台。Industry-first professional AI Agent platform for controllable film & video production. From shorts to live-action with Hollywood-standard workflows.

opencode. The open source coding agent.

awesome-nanobanana-pro. 🚀 An awesome list of curated Nano Banana pro prompts and examples. Your go-to resource for mastering prompt engineering and exploring the creative potential of the Nano banana pro(Nano banana 2) AI image model.

FastGen. NVIDIA FastGen: Fast Generation from Diffusion Models

openclaw. Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

x-algorithm. Algorithm powering the For You feed on X

Engram. Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models

GLM-Image. GLM-Image: Auto-regressive for Dense-knowledge and High-fidelity Image Generation.

Awesome-RL-for-Video-Generation. A curated list of papers on reinforcement learning for video generation

TurboDiffusion. TurboDiffusion: 100–200× Acceleration for Video Diffusion Models

VTP. [ECCV 2026] Towards Scalable Pre-training of Visual Tokenizers for Generation

Wan-Move. [NeurIPS 2025] Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance

FinRL. FinRL®: Financial Reinforcement Learning. 🔥

bitcoin-arbitrage. Bitcoin arbitrage - opportunity detector

gated_attention. The official implementation for [NeurIPS2025 Oral] Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free

Z-Image. Python

RepTok. [ICLR 2026] Adapting Self-Supervised Representations as a Latent Space for Efficient Generation

GenExam. [ICML 2026] GenExam: A Multidisciplinary Text-to-Image Exam

MiMo-Embodied. MiMo-Embodied

sam-3d-objects. SAM 3D Objects

sam3. The repository provides code for running inference and finetuning with the Meta Segment Anything Model 3 (SAM 3), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.

flame. 🔥 A minimal training framework for scaling FLA models

JiT. PyTorch implementation of JiT https://arxiv.org/abs/2511.13720

VideoREPA. [NeurIPS 2025] VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models

lejepa. Python

nvg. [ICLR 2026] Code for our paper "Next Visual Granularity Generation".

flash-moba. C++

flashrnn. FlashRNN - Fast RNN Kernels with I/O Awareness

ml-atoken. Jupyter Notebook

omnilingual-asr. Omnilingual ASR Open-Source Multilingual SpeechRecognition for 1600+ Languages

InfinityStar. [NeurIPS 2025 Oral]Infinity⭐️: Unified Spacetime AutoRegressive Modeling for Visual Generation

cat-transformer. Compress and Attend Transformers (CATs) 😸

PyPortfolioOpt. Financial portfolio optimization in python, including classical efficient frontier, Black-Litterman, Hierarchical Risk Parity

adata. 免费开源A股量化交易数据库； 专注A股，专注量化，向阳而生； 开放、纯净、持续、为Ai(爱)发电。为个人量化交易而生，保卫3000点，珍惜底部机会......【股票数据，股票行情数据，股票量化数据，股票交易数据，k线行情数据，股票概念数据，股票数据接口，行情数据接口，量化交易数据】【多数据源融合，动态设置代理，保障数据高可用性】

DanceGRPO. An official implementation of DanceGRPO: Unleashing GRPO on Visual Generation

Kimi-Linear.

Awesome-Visual-Generation-Alignment-Survey. A survey for visual generation alignment

tinker-cookbook. Post-training with Tinker

LongCat-Video. Python

LongLive. Long Video Gen Infrastructure

pico-banana-400k. Python

DreamOmni2. This project is the official implementation of 'DreamOmni2: Multimodal Instruction-based Editing and Generation (CVPR2026 Highlight)''

DeepSeek-OCR. Contexts Optical Compression

trading-signals. Technical indicators to run technical analysis with JavaScript & TypeScript. 📈

dexbotic. Dexbotic: Open-Source Vision-Language-Action Toolbox

Kandinsky-2. Kandinsky 2 — multilingual text2image latent diffusion model

edm2. EDM2 and Autoguidance -- Official PyTorch implementation

RAE. Official PyTorch Implementation of "Diffusion Transformers with Representation Autoencoders"

nanochat. The best ChatGPT that $100 can buy.

TinyRecursiveModels. Python

lmdeploy. LMDeploy is a toolkit for compressing, deploying, and serving LLMs.

PromptEnhancer. [CVPR 2026] PromptEnhancer is a prompt-rewriting tool, refining prompts into clearer, structured versions for better image generation.

Baichuan-Omni-1.5. Python

HuMo. HuMo: Human-Centric Video Generation via Collaborative Multi-Modal Conditioning

RobusTok. Image Tokenizer Needs Post-Training

Awesome-Nano-Banana-images. A curated collection of fun and creative examples generated with Nano Banana & Nano Banana Pro🍌, Gemini-2.5-flash-image based model. We also release Nano-consistent-150K openly to support the community's development of image generation and unified models(click to website to see our blog)

Waver. Industry-level video foundation model for unified Text-to-Video (T2V) and Image-to-Video (I2V) generation.

FastVGGT. [ICLR 2026] FastVGGT: Fast Visual Geometry Transformer

USO. [CVPR 2026] 🔥🔥 Official Repo of USO: Unified Style and Subject-Driven Generation via Disentangled and Reward Learning

RLinf. RLinf: Reinforcement Learning Infrastructure for Embodied and Agentic AI

HQ-Edit. [ICLR 2025] HQ-Edit: A High-Quality and High-Coverage Dataset for General Image Editing

ImageFolder. High-performance Image Tokenizers for VAR and AR

AttentionEngine. Python

dinov3. Reference PyTorch implementation and models for DINOv3

VeOmni. VeOmni: Scaling Any Modality Model Training with Model-Centric Distributed Recipe Zoo

DreamVVT. DreamVVT: Mastering Realistic Video Virtual Try-On in the Wild via a Stage-Wise Diffusion Transformer Framework

gpt-oss. gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI

Qwen-Image. Qwen-Image is a powerful image generation foundation model capable of complex text rendering and precise image editing.

DispLoss. Python

Wan2.2. Wan: Open and Advanced Large-Scale Video Generative Models

Pi3. [ICLR 2026] π^3: Permutation-Equivariant Visual Geometry Learning

hnet. H-Net: Hierarchical Network with Dynamic Chunking

Kimi-K2. Kimi K2 is the large language model series developed by Moonshot AI team

REG. [NeurIPS 2025 Oral] Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think

trae-agent. Trae Agent is an LLM-based agent for general purpose software engineering tasks.

perception_models. State-of-the-art Image & Video CLIP, Multimodal Large Language Models, and More!

HeavyBall. Efficient optimizers

lightning-thunder. PyTorch compiler that accelerates training and inference. Get built-in optimizations for performance, memory, parallelism, and easily write your own.

REPA-E. [ICCV 2025] Official implementation of the paper: REPA-E: Unlocking VAE for End-to-End Tuning of Latent Diffusion Transformers

LoRAEdit. We achieves high-quality first-frame guided video editing given a reference image, while maintaining flexibility for incorporating additional reference conditions.

vjepa2. PyTorch code and models for VJEPA2 self-supervised learning from video.

SeedVR. Repo for SeedVR2 (ICLR2026) & SeedVR (CVPR2025 Highlight)

MiniMax-M1. MiniMax-M1, the world's first open-weight, large-scale hybrid-attention reasoning model.

flux. Official inference repo for FLUX.1 models

OmniEdit. Official Repo for Paper "OmniEdit: Building Image Editing Generalist Models Through Specialist Supervision" [ICLR2025]

zoology. Understand and test language model architectures on synthetic tasks.

Dream. Dream 7B, a large diffusion language model

D-AR. the official repo for "D-AR: Diffusion via Autoregressive Models"

linear-attention-and-beyond-slides. TeX

BLIP3o. Official implementation of BLIP3o-Series

LaCT. Code release for paper "Test-Time Training Done Right"

DetailFlow. 🔥 Official impl. of "DetailFlow: 1D Coarse-to-Fine Autoregressive Image Generation via Next-Detail Prediction"

RF-Solver-Edit. [🚀ICML 2025] "Taming Rectified Flow for Inversion and Editing" Using FLUX and HunyuanVideo for image and video editing!

Bagel. Open-source unified multimodal model

3DDFA-V3. The official implementation of 3DDFA_V3 in CVPR2024 (Highlight).

deer-flow. An open-source long-horizon SuperAgent harness that researches, codes, and creates. With the help of sandboxes, memories, tools, skill, subagents and message gateway, it handles different levels of tasks that could take minutes to hours.

genesis-world. Simulation platform for general-purpose robotics & embodied AI learning.

SuperEdit. [ICCV 2025] Code & Data for: SuperEdit - Rectifying and Facilitating Supervision for Instruction-Based Image Editing

Awesome-FLUX-DiT. A collection of diffusion models based on FLUX/DiT for image/video generation, editing, reconstruction, inpainting .etc.

ICEdit. [NeurIPS 2025] Image editing is worth a single LoRA! 0.1% training data for fantastic image editing! Surpasses GPT-4o in ID persistence~ MoE ckpt released! Only 4GB VRAM is enough to run!

UniAnimate-DiT. UniAnimate-DiT: Human Image Animation with Large-Scale Video Diffusion Transformer

Step1X-Edit. A SOTA open-source image editing model, which aims to provide comparable performance against the closed-source models like GPT-4o and Gemini 2 Flash.

DreamO. [SIGGRAPH Asia 2025] DreamO: A Unified Framework for Image Customization

REPA. [ICLR'25 Oral] Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think

HiDream-I1. Python

MAGI-1. MAGI-1: Autoregressive Video Generation at Scale

MiniMax-01. The official repo of MiniMax-Text-01 and MiniMax-VL-01, large-language-model & vision-language-model based on Linear Attention

FramePack. Lets make video diffusion practical!

EasyControl. Implementation of "EasyControl: Adding Efficient and Flexible Control for Diffusion Transformer"(ICCV2025)

AnyEdit. 【CVPR 2025 Oral】Official Repo for Paper "AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea"

UnifyEdit. Tuning-Free Image Editing with Fidelity and Editability via Unified Latent Diffusion Model

ttt-video-dit. Official PyTorch implementation of One-Minute Video Generation with Test-Time Training

cuda-python. CUDA Python: Performance meets Productivity

Sana. SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer

LightningDiT. [CVPR 2025 Oral] Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models

BlaGPT. Experimental playground for benchmarking language model (LM) architectures, layers, and tricks on smaller datasets. Designed for flexible experimentation and exploration.

diffusion-4k. [CVPR 2025] Diffusion-4K: Ultra-High-Resolution Image Synthesis with Latent Diffusion Models

Wan2.1. Wan: Open and Advanced Large-Scale Video Generative Models

InfiniteYou. 🔥 [ICCV 2025 Highlight] InfiniteYou: Flexible Photo Recrafting While Preserving Your Identity

native-sparse-attention. 🐳 Efficient Triton implementations for "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention"

vggt. [CVPR 2025 Best Paper Award] VGGT: Visual Geometry Grounded Transformer

EquiTriton. EquiTriton is a project that seeks to implement high-performance kernels for commonly used building blocks in equivariant neural networks, enabling compute efficient training and inference.

DeepGEMM. DeepGEMM: clean and efficient BLAS kernel library on GPU

OpenManus. No fortress, purely open ground. OpenManus is Coming.

ktransformers. A Flexible Framework for Experiencing Heterogeneous LLM Inference/Fine-tune Optimizations

FLASHNN. Python

triton-index. Cataloging released Triton kernels.

attorch. A subset of PyTorch's neural network modules, written in Python using OpenAI's Triton.

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

xAR. This repository includes the official implementation of our paper "Beyond Next-Token: Next-X Prediction for Autoregressive Visual Generation"

DoRA. [ICML2024 (Oral)] Official PyTorch implementation of DoRA: Weight-Decomposed Low-Rank Adaptation

llm-foundry. LLM training code for Databricks foundation models

DeepEP. DeepEP: an efficient expert-parallel communication library

DeepClaude. Unleash Next-Level AI! 🚀 💻 Code Generation: DeepSeek r1 + Claude 3.7 Sonnet - Unparalleled Performance! 📝 Content Creation: DeepSeek r1 + Gemini 2.5 Pro - Superior Quality! 🔌 OpenAI-Compatible. 🌊 Streaming & Non-Streaming Support. ✨ Experience the Future of AI – Today! Click to Try Now! ✨

Moonlight. Muon is Scalable for LLM Training

FlashMLA. FlashMLA: Efficient Multi-head Latent Attention Kernels

Time-Series-Works-Conferences. Time-Series Work Summary in CS Top Conferences (NIPS, ICML, ICLR, KDD, AAAI, WWW, IJCAI, CIKM, ICDM, ICDE, etc.)

GatedDeltaNet. [ICLR 2025] Official PyTorch Implementation of Gated Delta Networks: Improving Mamba2 with Delta Rule

COAT. [ICLR 2025] COAT: Compressing Optimizer States and Activation for Memory-Efficient FP8 Training

Step-Video-T2V. Python

Qwen3. Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud.

MoBA. MoBA: Mixture of Block Attention for Long-Context LLMs

tritonbench. Tritonbench is a collection of PyTorch custom operators with example inputs to measure their performance.

imagesize_py. Python

unsloth. Unsloth is a local UI for training and running Kimi K3, Gemma 4, Qwen3.6, DeepSeek, GLM and other models.

TinyZero. Minimal reproduction of DeepSeek R1-Zero

FlagAttention. A collection of memory efficient attention operators implemented in the Triton language.

extension-cpp. C++ extensions in PyTorch

FlagGems. FlagGems is an operator library for large language models implemented in the Triton Language.

Liger-Kernel. Efficient Triton Kernels for LLM Training

DailyArXiv. Daily ArXiv Papers.

Qwen3-VL. Qwen3-VL is the multimodal large language model series developed by Qwen team, Alibaba Cloud.

Hunyuan3D-2. High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.

Kimi-k1.5.

DeepSeek-R1.

cosmos. NVIDIA Cosmos is an open platform of world models, datasets, and tools that enables developers to build Physical AI for robots, autonomous vehicles, smart infrastructure, and more.

xDiT. xDiT: A Scalable Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelism

flash-linear-attention. 🚀 Efficient implementations for emerging model architectures

micro_diffusion. Official repository for our work on micro-budget training of large-scale diffusion models.

DeepSeek-V3. Python

VideoVAEPlus. [ICCV 2025] VideoVAE+: Large Motion Video Autoencoding with Cross-modal Video VAE

Infinity. [CVPR 2025 Oral]Infinity ∞ : Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis

Ruyi-Models. Python

LlamaFactory. Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

VAR. [NeurIPS 2024 Best Paper Award][GPT beats diffusion🔥] [scaling laws in visual generation📈] Official impl. of "Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction". An *ultra-simple, user-friendly yet state-of-the-art* codebase for autoregressive image generation!

MARS. The official implementation of MARS: Unleashing the Power of Variance Reduction for Training Large Models

ai-toolkit. The ultimate training toolkit for finetuning diffusion models

Janus. Janus-Series: Unified Multimodal Understanding and Generation Models

addit. Python

RAG-Diffusion. [ICCV 2025] Region-Aware Text-to-Image Generation via Hard Binding and Soft Refinement 🔥

alphafold3. AlphaFold 3 inference pipeline.

EmbodiedCity. Python

mirage. Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

xformers. Hackable and optimized Transformers building blocks, supporting a composable construction.

Awesome-Mamba-Papers. Awesome Papers related to Mamba.

sapiens. High-resolution models for human tasks.

ControlNeXt. Controllable video and image Generation, SVD, Animate Anyone, ControlNet, ControlNeXt, LoRA

mattermost. Mattermost is an open source platform for secure collaboration across the entire software development lifecycle..

IDM-VTON. [ECCV2024] IDM-VTON : Improving Diffusion Models for Authentic Virtual Try-on in the Wild

sam2. The repository provides code for running inference with the Meta Segment Anything Model 2 (SAM 2), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.

mem0. Universal memory layer for AI Agents

mplfinance. Financial Markets Data Visualization using Matplotlib

iTransformer. Official implementation for "iTransformer: Inverted Transformers Are Effective for Time Series Forecasting" (ICLR 2024 Spotlight)

X-Pose. [ECCV 2024] Official implementation of the paper "X-Pose: Detecting Any Keypoints"

LivePortrait. Bring portraits to life!