DefTruth / ffpa-attn-mma
📚FFPA(Split-D): Yet another Faster Flash Prefill Attention with O(1) GPU SRAM complexity for headdim > 256, ~2x↑🎉vs SDPA EA.
☆147Updated this week
Alternatives and similar repositories for ffpa-attn-mma:
Users that are interested in ffpa-attn-mma are comparing it to the libraries listed below
- ☆74Updated 3 months ago
- Implement Flash Attention using Cute.☆71Updated 3 months ago
- Examples of CUDA implementations by Cutlass CuTe☆145Updated last month
- ☆87Updated 6 months ago
- ⚡️Write HGEMM from scratch using Tensor Cores with WMMA, MMA and CuTe API, Achieve Peak⚡️ Performance.☆59Updated 2 weeks ago
- We invite you to visit and follow our new repository at https://github.com/microsoft/TileFusion. TiledCUDA is a highly efficient kernel …☆178Updated last month
- Puzzles for learning Triton, play it with minimal environment configuration!☆262Updated 3 months ago
- Performance of the C++ interface of flash attention and flash attention v2 in large language model (LLM) inference scenarios.☆35Updated 3 weeks ago
- flash attention tutorial written in python, triton, cuda, cutlass☆299Updated 2 months ago
- Standalone Flash Attention v2 kernel without libtorch dependency☆106Updated 6 months ago
- A Easy-to-understand TensorOp Matmul Tutorial☆327Updated 6 months ago
- QQQ is an innovative and hardware-optimized W4A8 quantization solution for LLMs.☆108Updated last week
- A collection of memory efficient attention operators implemented in the Triton language.☆253Updated 9 months ago
- ☆87Updated last week
- ☆145Updated 2 months ago
- ☆46Updated 2 months ago
- 使用 cutlass 仓库在 ada 架构上实现 fp8 的 flash attention☆60Updated 7 months ago
- High-speed GEMV kernels, at most 2.7x speedup compared to pytorch baseline.☆100Updated 8 months ago
- ☆113Updated last year
- ☆110Updated 3 months ago
- A standalone GEMM kernel for fp16 activation and quantized weight, extracted from FasterTransformer☆89Updated 3 weeks ago
- Tutorials for writing high-performance GPU operators in AI frameworks.☆129Updated last year
- Decoding Attention is specially optimized for MHA, MQA, GQA and MLA using CUDA core for the decoding stage of LLM inference.☆35Updated 2 weeks ago
- Several optimization methods of half-precision general matrix vector multiplication (HGEMV) using CUDA core.☆57Updated 6 months ago
- ☆191Updated 8 months ago
- ☆139Updated 11 months ago
- Benchmark code for the "Online normalizer calculation for softmax" paper☆85Updated 6 years ago
- ☆87Updated last year
- ☆159Updated last year