thu-ml / SageAttention
Quantized Attention that achieves speedups of 2.1x and 2.7x compared to FlashAttention2 and xformers, respectively, without lossing end-to-end metrics across various models.
☆403Updated this week
Related projects ⓘ
Alternatives and complementary repositories for SageAttention
- xDiT: A Scalable Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelism☆714Updated this week
- Model Compression Toolbox for Large Language Models and Diffusion Models☆222Updated last week
- [CVPR 2024 Highlight] DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models☆590Updated 2 weeks ago
- ☆100Updated last month
- SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models☆327Updated this week
- 📒A small curated list of Awesome Diffusion Inference Papers with codes.☆96Updated this week
- USP: Unified (a.k.a. Hybrid, 2D) Sequence Parallel Attention for Long Context Transformers Model Training and Inference☆357Updated this week
- DuoAttention: Efficient Long-Context LLM Inference with Retrieval and Streaming Heads☆373Updated 3 weeks ago
- [ICML 2024] Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference☆202Updated 2 weeks ago
- QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving☆443Updated last week
- Ring attention implementation with flash attention☆585Updated last week
- An efficient GPU support for LLM inference with x-bit quantization (e.g. FP6,FP5).☆208Updated 3 weeks ago
- Code for Neurips24 paper: QuaRot, an end-to-end 4-bit inference of large language models.☆284Updated 3 months ago
- [ICML 2024] KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache☆241Updated last month
- [MLSys'24] Atom: Low-bit Quantization for Efficient and Accurate LLM Serving☆278Updated 4 months ago
- [COLM 2024] TriForce: Lossless Acceleration of Long Sequence Generation with Hierarchical Speculative Decoding☆230Updated 2 months ago
- A collection of memory efficient attention operators implemented in the Triton language.☆219Updated 5 months ago
- FlagGems is an operator library for large language models implemented in Triton Language.☆342Updated this week
- ☆188Updated 6 months ago
- Dynamic Memory Management for Serving LLMs without PagedAttention☆238Updated last week
- [NeurIPS'23] H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models.☆391Updated 3 months ago
- flash attention tutorial written in python, triton, cuda, cutlass☆202Updated 5 months ago
- [EMNLP 2024 Industry Track] This is the official PyTorch implementation of "LLMC: Benchmarking Large Language Model Quantization with a V…☆322Updated this week
- Patch convolution to avoid large GPU memory usage of Conv2D☆79Updated 5 months ago
- QQQ is an innovative and hardware-optimized W4A8 quantization solution for LLMs.☆87Updated last month
- [ICCV 2023] Q-Diffusion: Quantizing Diffusion Models.☆331Updated 8 months ago
- A parallelism VAE avoids OOM for high resolution image generation☆40Updated last month
- ☆289Updated 7 months ago
- [WIP] Context parallel attention that works with torch.compile☆49Updated this week
- BitBLAS is a library to support mixed-precision matrix multiplications, especially for quantized LLM deployment.☆420Updated this week