File tree Expand file tree Collapse file tree 2 files changed +20
-26
lines changed Expand file tree Collapse file tree 2 files changed +20
-26
lines changed Original file line number Diff line number Diff line change @@ -118,32 +118,29 @@ The following table shows the backward pass performance comparison between Flash
118118
119119## Installation
120120
121- ### Prerequisites
121+ ### Requirements
122122
123- - ** Python** : 3.8 or later
124- - ** PyTorch** : 2.0.0 or later
125- - ** CUDA** : 11.8 or later
123+ - ** Linux** : Ubuntu 22.04 or later
126124- ** NVIDIA GPU** : Compute Capability 8.0 or higher
127125- ** C++ Compiler** : GCC 7+
126+ - ** CUDA** : 11.8 or later
127+ - ** Python** : 3.9 or later
128+ - ** PyTorch** : 2.5.1 or later
128129
129- ### CUDA Environment Setup
130+ ### Install
130131
131- Ensure your CUDA environment is properly configured :
132+ You can install Flash-DMA via pre-compiled wheels :
132133
133134``` bash
134- # Check CUDA installation
135- nvcc --version
136-
137- # Set CUDA_HOME if needed
138- export CUDA_HOME=/usr/local/cuda
135+ pip install flash-dmattn --no-build-isolation
139136```
140137
141- ### Install from Source
138+ Alternatively, you can compile and install from source:
142139
143140``` bash
144141git clone https://github.com/SmallDoges/flash-dmattn.git
145142cd flash-dmattn
146- MAX_JOBS=4 pip install . --no-build-isolation
143+ pip install . --no-build-isolation
147144```
148145
149146
Original file line number Diff line number Diff line change @@ -118,32 +118,29 @@ Flash-DMA 是一个高性能的注意力实现,将 Flash Attention 的内存
118118
119119## 安装
120120
121- ### 先决条件
121+ ### 依赖
122122
123- - ** Python** : 3.8 或更高版本
124- - ** PyTorch** : 2.0.0 或更高版本
125- - ** CUDA** : 11.8 或更高版本
123+ - ** Linux** : Ubuntu 22.04 或更高版本
126124- ** NVIDIA GPU** : 计算能力 8.0 或更高
127125- ** C++ 编译器** : GCC 7+
126+ - ** CUDA** : 11.8 或更高版本
127+ - ** Python** : 3.9 或更高版本
128+ - ** PyTorch** : 2.5.1 或更高版本
128129
129- ### CUDA 环境设置
130+ ### 安装
130131
131- 确保您的 CUDA 环境已正确配置 :
132+ 您可以通过预编译的轮子安装 Flash-DMA :
132133
133134``` bash
134- # 检查 CUDA 安装
135- nvcc --version
136-
137- # 如需要,设置 CUDA_HOME
138- export CUDA_HOME=/usr/local/cuda
135+ pip install flash-dmattn --no-build-isolation
139136```
140137
141- ### 从源码安装
138+ 或者,您可以从源代码编译和安装:
142139
143140``` bash
144141git clone https://github.com/SmallDoges/flash-dmattn.git
145142cd flash-dmattn
146- MAX_JOBS=4 pip install . --no-build-isolation
143+ pip install . --no-build-isolation
147144```
148145
149146
You can’t perform that action at this time.
0 commit comments