Helpful tools and examples for working with flex-attention
☆1,140Feb 8, 2026Updated last month
Alternatives and similar repositories for attention-gym
Users that are interested in attention-gym are comparing it to the libraries listed below
Sorting:
- A PyTorch native platform for training generative AI models☆5,111Updated this week
- 🚀 Efficient implementations of state-of-the-art linear attention models☆4,474Updated this week
- Tile primitives for speedy kernels☆3,202Feb 24, 2026Updated last week
- Ring attention implementation with flash attention☆987Sep 10, 2025Updated 5 months ago
- PyTorch native quantization and sparsity for training and inference☆2,722Updated this week
- Efficient Triton Kernels for LLM Training☆6,189Updated this week
- FlashInfer: Kernel Library for LLM Serving☆5,101Updated this week
- Distributed Compiler based on Triton for Parallel Systems☆1,371Feb 13, 2026Updated 3 weeks ago
- A subset of PyTorch's neural network modules, written in Python using OpenAI's Triton.☆595Aug 12, 2025Updated 6 months ago
- A Python-embedded DSL that makes it easy to write fast, scalable ML kernels with minimal boilerplate.☆774Updated this week
- 🐳 Efficient Triton implementations for "Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention"☆974Feb 5, 2026Updated last month
- A sparse attention kernel supporting mix sparse patterns☆472Jan 18, 2026Updated last month
- A Distributed Attention Towards Linear Scalability for Ultra-Long Context, Heterogeneous Data Training☆659Updated this week
- A library for accelerating Transformer models on NVIDIA GPUs, including using 8-bit and 4-bit floating point (FP8 and FP4) precision on H…☆3,187Updated this week
- Fast and memory-efficient exact attention☆22,460Updated this week
- Minimalistic large language model 3D-parallelism training☆2,588Feb 19, 2026Updated 2 weeks ago
- Applied AI experiments and examples for PyTorch☆319Aug 22, 2025Updated 6 months ago
- Hackable and optimized Transformers building blocks, supporting a composable construction.☆10,356Feb 20, 2026Updated 2 weeks ago
- A Quirky Assortment of CuTe Kernels☆838Updated this week
- Large Context Attention☆769Oct 13, 2025Updated 4 months ago
- USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference☆644Jan 15, 2026Updated last month
- Puzzles for learning Triton☆2,324Nov 18, 2024Updated last year
- Using FlexAttention to compute attention with different masking patterns☆47Sep 22, 2024Updated last year
- Tritonbench is a collection of PyTorch custom operators with example inputs to measure their performance.☆327Mar 1, 2026Updated last week
- 🔥 A minimal training framework for scaling FLA models☆352Nov 15, 2025Updated 3 months ago
- [ICML2025] SpargeAttention: A training-free sparse attention that accelerates any model inference.☆954Feb 25, 2026Updated last week
- [ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-t…☆3,192Jan 17, 2026Updated last month
- [ICML 2025] XAttention: Block Sparse Attention with Antidiagonal Scoring☆269Jul 6, 2025Updated 8 months ago
- Mirage Persistent Kernel: Compiling LLMs into a MegaKernel☆2,148Feb 23, 2026Updated 2 weeks ago
- Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.☆6,185Aug 22, 2025Updated 6 months ago
- FlexAttention w/ FlashAttention3 Support☆27Oct 5, 2024Updated last year
- Flash-Muon: An Efficient Implementation of Muon Optimizer☆239Jun 15, 2025Updated 8 months ago
- A fast communication-overlapping library for tensor/expert parallelism on GPUs.☆1,264Aug 28, 2025Updated 6 months ago
- Domain-specific language designed to streamline the development of high-performance GPU/CPU/Accelerators kernels☆5,330Updated this week
- ☆136May 29, 2025Updated 9 months ago
- Schedule-Free Optimization in PyTorch☆2,262May 21, 2025Updated 9 months ago
- ☆105Nov 7, 2024Updated last year
- FP16xINT4 LLM inference kernel that can achieve near-ideal ~4x speedups up to medium batchsizes of 16-32 tokens.☆1,035Sep 4, 2024Updated last year
- Fault tolerance for PyTorch (HSDP, LocalSGD, DiLoCo, Streaming DiLoCo)☆481Updated this week