Alternatives and similar repositories for flash-attention

Users that are interested in flash-attention are comparing it to the libraries listed below

• InternLM / turbomind
  ☆96Updated 10 months ago
• infinigence / Semi-PD
  A prefill & decode disaggregated LLM serving framework with shared GPU memory and fine-grained compute isolation.
  ☆123Updated last month
• AlibabaPAI / FLASHNN
  ☆105Updated last year
• infinigence / FlashOverlap
  A lightweight design for computation-communication overlap.
  ☆219Updated 3 weeks ago
• DeepLink-org / DLSlime
  DLSlime: Flexible & Efficient Heterogeneous Transfer Toolkit
  ☆92Updated 2 weeks ago
• microsoft / vattention
  Dynamic Memory Management for Serving LLMs without PagedAttention
  ☆458Updated 8 months ago
• xlite-dev / HGEMM
  ⚡️Write HGEMM from scratch using Tensor Cores with WMMA, MMA and CuTe API, Achieve Peak⚡️ Performance.
  ☆148Updated 9 months ago
• madsys-dev / deepseekv2-profile
  ☆155Updated 11 months ago
• ademeure / DeeperGEMM
  DeeperGEMM: crazy optimized version
  ☆73Updated 9 months ago
• flagos-ai / FlagCX
  FlagCX is a scalable and adaptive cross-chip communication library.
  ☆172Updated last week
• stepfun-ai / StepMesh
  ☆342Updated 2 weeks ago
• inclusionAI / asystem-amem
  A NCCL extension library, designed to efficiently offload GPU memory allocated by the NCCL communication library.
  ☆91Updated last month
• feifeibear / LLMRoofline
  Compare different hardware platforms via the Roofline Model for LLM inference tasks.
  ☆120Updated last year
• bytedance / xpu-perf
  AI Accelerator Benchmark focuses on evaluating AI Accelerators from a practical production perspective, including the ease of use and ver…
  ☆298Updated this week
• LLMServe / SwiftTransformer
  High performance Transformer implementation in C++.
  ☆151Updated last year
• antgroup / DeepXTrace
  DeepXTrace is a lightweight tool for precisely diagnosing slow ranks in DeepEP-based environments.
  ☆92Updated 3 weeks ago
• AlibabaResearch / flash-llm
  Flash-LLM: Enabling Cost-Effective and Highly-Efficient Large Generative Model Inference with Unstructured Sparsity
  ☆233Updated 2 years ago
• flagos-ai / FlagAttention
  A collection of memory efficient attention operators implemented in the Triton language.
  ☆287Updated last year
• usyd-fsalab / fp6_llm
  An efficient GPU support for LLM inference with x-bit quantization (e.g. FP6,FP5).
  ☆276Updated 6 months ago
• PKU-SEC-Lab / HybriMoE
  [DAC'25] Official implement of "HybriMoE: Hybrid CPU-GPU Scheduling and Cache Management for Efficient MoE Inference"
  ☆101Updated last month
• microsoft / sarathi-serve
  A low-latency & high-throughput serving engine for LLMs
  ☆470Updated last month
• KuangjuX / NVSHMEM-Tutorial
  NVSHMEM‑Tutorial: Build a DeepEP‑like GPU Buffer
  ☆161Updated 4 months ago
• Ascend / triton-ascend
  Triton adapter for Ascend. Mirror of https://gitee.com/ascend/triton-ascend
  ☆107Updated this week
• flexflow / flexflow-serve
  FlexFlow Serve: Low-Latency, High-Performance LLM Serving
  ☆71Updated 4 months ago
• meta-pytorch / KernelAgent
  Autonomous GPU Kernel Generation via Deep Agents
  ☆233Updated this week
• Bruce-Lee-LY / flash_attention_inference
  Performance of the C++ interface of flash attention and flash attention v2 in large language model (LLM) inference scenarios.
  ☆44Updated 11 months ago
• microsoft / AttentionEngine
  ☆118Updated 8 months ago
• xlite-dev / ffpa-attn
  🤖FFPA: Extend FlashAttention-2 with Split-D, ~O(1) SRAM complexity for large headdim, 1.8x~3x↑🎉 vs SDPA EA.
  ☆250Updated last week
• triton-lang / kernels
  ☆104Updated last year
• microsoft / tokenweave
  Accepted to MLSys 2026
  ☆70Updated 2 weeks ago