Bruce-Lee-LY / decoding_attentionLinks
Decoding Attention is specially optimized for MHA, MQA, GQA and MLA using CUDA core for the decoding stage of LLM inference.
☆44Updated 3 months ago
Alternatives and similar repositories for decoding_attention
Users that are interested in decoding_attention are comparing it to the libraries listed below
Sorting:
- Performance of the C++ interface of flash attention and flash attention v2 in large language model (LLM) inference scenarios.☆40Updated 7 months ago
- ☆64Updated 5 months ago
- Multiple GEMM operators are constructed with cutlass to support LLM inference.☆19Updated 2 months ago
- ⚡️Write HGEMM from scratch using Tensor Cores with WMMA, MMA and CuTe API, Achieve Peak⚡️ Performance.☆118Updated 4 months ago
- Implement Flash Attention using Cute.☆95Updated 9 months ago
- Quantized Attention on GPU☆44Updated 10 months ago
- Standalone Flash Attention v2 kernel without libtorch dependency☆110Updated last year
- ☆98Updated last year
- A GPU-optimized system for efficient long-context LLMs decoding with low-bit KV cache.☆60Updated last month
- ☆95Updated 6 months ago
- ☆32Updated 8 months ago
- A standalone GEMM kernel for fp16 activation and quantized weight, extracted from FasterTransformer☆94Updated 2 weeks ago
- ☆98Updated 4 months ago
- 使用 cutlass 仓库在 ada 架构上实现 fp8 的 flash attention☆75Updated last year
- ☆82Updated 8 months ago
- A practical way of learning Swizzle☆28Updated 8 months ago
- GPTQ inference TVM kernel☆40Updated last year
- ☆50Updated 4 months ago
- ☆12Updated 8 months ago
- ☆112Updated last month
- ☆106Updated 4 months ago
- DeeperGEMM: crazy optimized version☆71Updated 4 months ago
- 使用 CUDA C++ 实现的 llama 模型推理框架☆62Updated 10 months ago
- Persistent dense gemm for Hopper in `CuTeDSL`☆15Updated last month
- llama INT4 cuda inference with AWQ☆55Updated 8 months ago
- SpInfer: Leveraging Low-Level Sparsity for Efficient Large Language Model Inference on GPUs☆58Updated 6 months ago
- 🤖FFPA: Extend FlashAttention-2 with Split-D, ~O(1) SRAM complexity for large headdim, 1.8x~3x↑🎉 vs SDPA EA.☆219Updated last month
- Several optimization methods of half-precision general matrix vector multiplication (HGEMV) using CUDA core.☆67Updated last year
- ☆13Updated 6 months ago
- TileFusion is an experimental C++ macro kernel template library that elevates the abstraction level in CUDA C for tile processing.☆97Updated 3 months ago