diff --git a/doc/release-note.md b/doc/release-note.md index e9372883..72f69da0 100644 --- a/doc/release-note.md +++ b/doc/release-note.md @@ -1,3 +1,24 @@ +## v0.1.1 + +### Distributed + +- Broadcast data-parallel parameters before training. + +### Megatron adaption + +- Initialize `FMoELinear` parameters using different seed in model parallel even using the same random seed in megatron. +- Use proper comm for mp and dp. + +### Transformer-XL example + +- Improve scripts. + +### Misc + +- Logo and slack workspace link. +- Document in Chinese. +- Figures to explain how FastMoE works. + ## v0.1.0 ### Functions diff --git a/setup.py b/setup.py index ddd3a15a..0929cac9 100644 --- a/setup.py +++ b/setup.py @@ -15,9 +15,9 @@ if __name__ == '__main__': setuptools.setup( - name='fmoe', - version='0.1.0', - description='An efficient Mixture-of-Experts impl. for PyTorch', + name='fastmoe', + version='0.1.1', + description='An efficient Mixture-of-Experts system for PyTorch', author='Jiaao He, Jiezhong Qiu and Aohan Zeng', author_email='hja20@mails.tsinghua.edu.cn', license='Apache-2',