Repositories list

• vlash

  Public

  mit-han-lab/vlash’s past year of commit activity
  Real-Time VLAs via Future-state-aware Asynchronous Inference.

  Python

  •

  Apache License 2.0

  •8•200•3•0•Updated Dec 8, 2025Dec 8, 2025

• fastrl

  Public

  mit-han-lab/fastrl’s past year of commit activity
  [ASPLOS'26] Taming the Long-Tail: Efficient Reasoning RL Training with Adaptive Drafter

  Python

  •

  Apache License 2.0

  •8•100•2•0•Updated Dec 5, 2025Dec 5, 2025

• fouroversix

  Public
  Code for the paper “Four Over Six: More Accurate NVFP4 Quantization with Adaptive Block Scaling”

  Python

  •

  MIT License

  •1•61•1•0•Updated Dec 4, 2025Dec 4, 2025

• flash-moba

  Public
  C++

  •

  BSD 3-Clause "New" or "Revised" License

  •6•203•4•0•Updated Nov 19, 2025Nov 19, 2025

• radial-attention

  Public

  mit-han-lab/radial-attention’s past year of commit activity
  [NeurIPS 2025] Radial Attention: O(nlogn) Sparse Attention with Energy Decay for Long Video Generation

  wanmochidiffusion-models+ 3sparse-attentionefficientmlhunyuan-video

  Python

  •

  Apache License 2.0

  •29•562•15•1•Updated Nov 11, 2025Nov 11, 2025

• torchquantum

  Public
  A PyTorch-based framework for Quantum Classical Simulation, Quantum Machine Learning, Quantum Neural Networks, Parameterized Quantum Circuits with support for easy deployments on real quantum computers.

  machine-learningsystemdeep-learning+ 10neural-networkquantumpytorchquantum-computingquantum-machine-learningquantum-simulationml-for-systems+ 3

  Jupyter Notebook

  •

  MIT License

  •242•1.6k•69•9•Updated Oct 28, 2025Oct 28, 2025

• streaming-vlm

  Public

  mit-han-lab/streaming-vlm’s past year of commit activity
  StreamingVLM: Real-Time Understanding for Infinite Video Streams

  Python

  •

  MIT License

  •49•753•13•0•Updated Oct 15, 2025Oct 15, 2025

• efficientvit

  Public

  mit-han-lab/efficientvit’s past year of commit activity
  Efficient vision foundation models for high-resolution generation and perception.

  imagenetsegmentationhigh-resolution+ 5vision-transformerefficientvitsegment-anythingdeep-compression-autoencoderefficient-diffusion-model

  Python

  •

  Apache License 2.0

  •231•3.2k•108•0•Updated Sep 5, 2025Sep 5, 2025

• llm-awq

  Public

  mit-han-lab/llm-awq’s past year of commit activity
  [MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration

  Python

  •

  MIT License

  •282•3.4k•169•10•Updated Jul 17, 2025Jul 17, 2025

• lpd

  Public
  Locality-aware Parallel Decoding for Efficient Autoregressive Image Generation

  accelerationimagenetimage-generation+ 3autoregressiveefficient-algorithmparallel-decoding

  Python

  •

  MIT License

  •6•80•1•0•Updated Jul 13, 2025Jul 13, 2025

• Quest

  Public
  [ICML 2024] Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference

  Cuda

  •

  MIT License

  •38•356•5•0•Updated Jul 10, 2025Jul 10, 2025

• x-attention

  Public
  [ICML 2025] XAttention: Block Sparse Attention with Antidiagonal Scoring

  Python

  •15•255•9•0•Updated Jul 6, 2025Jul 6, 2025

• vila-u

  Public
  [ICLR 2025] VILA-U: a Unified Foundation Model Integrating Visual Understanding and Generation

  Python

  •

  MIT License

  •17•411•18•0•Updated Apr 25, 2025Apr 25, 2025

• omniserve

  Public
  [MLSys'25] QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving; [MLSys'25] LServe: Efficient Long-sequence LLM Serving with Unified Sparse Attention

  C++

  •

  Apache License 2.0

  •54•790•46•4•Updated Mar 6, 2025Mar 6, 2025

• torchsparse

  Public
  [MICRO'23, MLSys'22] TorchSparse: Efficient Training and Inference Framework for Sparse Convolution on GPUs.

  accelerationpytorch

  Cuda

  •

  MIT License

  •179•1.4k•51•3•Updated Feb 24, 2025Feb 24, 2025

• VisCompare

  Public
  A WebUI for Side-by-Side Comparison of Media (Images/Videos) Across Multiple Folders

  pythonuivideo+ 5imageswebuidiffusion-modelsstreamlitgenerative-ai

  Python

  •

  Apache License 2.0

  •2•24•0•0•Updated Feb 21, 2025Feb 21, 2025

• Block-Sparse-Attention

  Public
  A sparse attention kernel supporting mix sparse patterns

  C++

  •

  BSD 3-Clause "New" or "Revised" License

  •34•401•10•1•Updated Feb 13, 2025Feb 13, 2025

• duo-attention

  Public
  [ICLR 2025] DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads

  Python

  •

  MIT License

  •36•510•14•1•Updated Feb 10, 2025Feb 10, 2025

• patch_conv

  Public

  mit-han-lab/patch_conv’s past year of commit activity
  Patch convolution to avoid large GPU memory usage of Conv2D

  Python

  •

  MIT License

  •7•93•2•1•Updated Jan 23, 2025Jan 23, 2025

• fastcomposer

  Public
  [IJCV] FastComposer: Tuning-Free Multi-Subject Image Generation with Localized Attention

  Python

  •

  MIT License

  •41•709•16•0•Updated Jan 10, 2025Jan 10, 2025

• sparserefine

  Public

  mit-han-lab/sparserefine’s past year of commit activity
  [ECCV 2024] SparseRefine: Sparse Refinement for Efficient High-Resolution Semantic Segmentation

  Python

  •

  MIT License

  •1•15•1•0•Updated Jan 10, 2025Jan 10, 2025

• distrifuser

  Public

  mit-han-lab/distrifuser’s past year of commit activity
  [CVPR 2024 Highlight] DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models

  accelerationparallelismgenerative-model+ 2diffusion-modelsgenerative-ai

  Python

  •

  MIT License

  •30•713•11•1•Updated Dec 2, 2024Dec 2, 2024

• tinyengine

  Public
  [NeurIPS 2020] MCUNet: Tiny Deep Learning on IoT Devices; [NeurIPS 2021] MCUNetV2: Memory-Efficient Patch-based Inference for Tiny Deep Learning; [NeurIPS 2022] MCUNetV3: On-Device Training Under 256KB Memory

  cmicrocontrollerdeep-learning+ 7cpppytorchcodegeneratorquantizationedge-computingneural-architecture-searchtinyml

  C

  •

  MIT License

  •155•910•35•1•Updated Nov 27, 2024Nov 27, 2024

• tinychat-tutorial

  Public

  mit-han-lab/tinychat-tutorial’s past year of commit activity
  C++

  •26•76•4•3•Updated Nov 5, 2024Nov 5, 2024

• hart

  Public

  mit-han-lab/hart’s past year of commit activity
  HART: Efficient Visual Generation with Hybrid Autoregressive Transformer

  Python

  •

  MIT License

  •41•644•19•2•Updated Oct 16, 2024Oct 16, 2024

• data-efficient-gans

  Public
  [NeurIPS 2020] Differentiable Augmentation for Data-Efficient GAN Training

  tensorflowpytorchgenerative-adversarial-network+ 4image-generationgansdata-efficientneurips-2020

  Python

  •

  BSD 2-Clause "Simplified" License

  •176•1.3k•28•0•Updated Sep 24, 2024Sep 24, 2024

• proxylessnas

  Public
  [ICLR 2019] ProxylessNAS: Direct Neural Architecture Search on Target Task and Hardware

  accelerationautomlspecialization+ 3efficient-modelon-device-aihardware-aware

  C++

  •

  MIT License

  •286•1.5k•0•2•Updated Aug 30, 2024Aug 30, 2024

• spatten

  Public
  [HPCA'21] SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning

  rtlattentionhardware-acceleration+ 2spinalhdlllm-inference

  Scala

  •

  MIT License

  •11•116•2•0•Updated Aug 27, 2024Aug 27, 2024

• bevfusion

  Public archive
  [ICRA'23] BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation

  camerapytorchlidar+ 4object-detectionsensor-fusionsemantic-segmentation3d-perception

  Python

  •

  Apache License 2.0

  •530•2.9k•0•0•Updated Jul 31, 2024Jul 31, 2024

• hardware-aware-transformers

  Public
  [ACL'20] HAT: Hardware-Aware Transformers for Efficient Natural Language Processing

  natural-language-processingmachine-translationtransformer+ 3specializationefficient-modelhardware-aware

  Python

  •

  Other

  •51•336•3•0•Updated Jul 14, 2024Jul 14, 2024