NVIDIA / online-softmaxLinks
Benchmark code for the "Online normalizer calculation for softmax" paper
☆94Updated 6 years ago
Alternatives and similar repositories for online-softmax
Users that are interested in online-softmax 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.☆37Updated 3 months ago
- Standalone Flash Attention v2 kernel without libtorch dependency☆109Updated 8 months ago
- 使用 cutlass 仓库在 ada 架构上实现 fp8 的 flash attention☆68Updated 9 months ago
- High-speed GEMV kernels, at most 2.7x speedup compared to pytorch baseline.☆109Updated 10 months ago
- ☆96Updated 8 months ago
- ☆73Updated 2 weeks ago
- A standalone GEMM kernel for fp16 activation and quantized weight, extracted from FasterTransformer☆92Updated last week
- Several optimization methods of half-precision general matrix vector multiplication (HGEMV) using CUDA core.☆62Updated 8 months ago
- ☆208Updated 10 months ago
- We invite you to visit and follow our new repository at https://github.com/microsoft/TileFusion. TiledCUDA is a highly efficient kernel …☆182Updated 4 months ago
- llama INT4 cuda inference with AWQ☆54Updated 4 months ago
- ⚡️Write HGEMM from scratch using Tensor Cores with WMMA, MMA and CuTe API, Achieve Peak⚡️ Performance.☆79Updated 3 weeks ago
- play gemm with tvm☆91Updated last year
- ☆86Updated 5 months ago
- ☆109Updated 3 weeks ago
- A lightweight design for computation-communication overlap.☆132Updated last month
- PyTorch bindings for CUTLASS grouped GEMM.☆93Updated last week
- Implement Flash Attention using Cute.☆85Updated 5 months ago
- ☆93Updated 2 months ago
- ☆59Updated last month
- 使用 cutlass 实现 flash-attention 精简版,具有教学意义☆41Updated 9 months ago
- ☆148Updated 4 months ago
- TileFusion is an experimental C++ macro kernel template library that elevates the abstraction level in CUDA C for tile processing.☆88Updated last week
- ☆73Updated 4 months ago
- 📚FFPA(Split-D): Extend FlashAttention with Split-D for large headdim, O(1) GPU SRAM complexity, 1.8x~3x↑🎉 faster than SDPA EA.☆184Updated 3 weeks ago
- Examples of CUDA implementations by Cutlass CuTe☆188Updated 4 months ago
- ☆33Updated last year
- ☆139Updated last year
- GPTQ inference TVM kernel☆40Updated last year
- ☆121Updated 6 months ago