Oneflow-Inc / libai

LiBai(李白): A Toolbox for Large-Scale Distributed Parallel Training
https://libai.readthedocs.io
Apache License 2.0
391 stars 55 forks source link

[MT5] exec_graph.cpp physical shape check failed. #405

Closed strint closed 1 year ago

strint commented 2 years ago

git branch:https://github.com/Oneflow-Inc/oneflow/pull/9245

安装方法: python3 -m pip install --pre oneflow -f https://staging.oneflow.info/branch/release/compile_cost_cnt/cu112

执行 mt5,报如下错误

image

是抱怨 op 的 infer shape 推理的 physical shape 和用 sbp 推理的 physical shape 不一致。

zipeng 在复现这个问题。

strint commented 2 years ago

这个问题 idea 小伙伴没有复现了,我们自己也没有复现。

后面可以复现了,再继续处理,先关闭。

strint commented 1 year ago

fix in:https://github.com/Oneflow-Inc/libai/issues/409