Unverified Commit 53844b65 authored by Rick Ho's avatar Rick Ho Committed by GitHub
Browse files

Merge pull request #136 from laekov/update-versions

Update version requirement in the documents
parents 53bac444 d9b6359d
......@@ -14,7 +14,11 @@ model for PyTorch.
### Prerequisites
PyTorch with CUDA is required. The repository is currently tested with PyTorch
v1.8.0 and CUDA 10, with designed compatibility to older versions.
v1.10.0 and CUDA 11.3, with designed compatibility to older and newer versions.
The minimum version of supported PyTorch is `1.7.2` with CUDA `10`. However,
there are a few known issues that requires manual modification of FastMoE's
code with specific older dependents.
If the distributed expert feature is enabled, NCCL with P2P communication
support, typically versions `>=2.7.5`, is needed.
......
......@@ -12,8 +12,11 @@ FastMoE 是一个易用且高效的基于 PyTorch 的 MoE 模型训练系统.
### 依赖
启用了 CUDA 的 PyTorch 是必要的. 当前版本的 FastMoE 在 PyTorch v1.8.0 和 CUDA 10
的平台上经过了测试. 本系统从设计上也支持更旧的 PyTorch 版本.
启用了 CUDA 的 PyTorch 是必要的. 当前版本的 FastMoE 在 PyTorch v1.10.0 和 CUDA
11 的平台上经过了测试. 本系统从设计上也支持更旧或更新的 PyTorch 版本.
已知最老的支持的版本是 PyTorch `1.7.0` 和 CUDA `10`,
但已知某些老版本可能需要修改 FastMoE 的代码以实现支持.
如果需要使能 FastMoE 模型并行特性, 那么支持点对点通信的 NCCL 库 (即不旧于
`2.7.5` 版本) 也是必需的.
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment