search

Oneflow-Inc / models

Models and examples built with OneFlow
Apache License 2.0
94 stars 37 forks source link

!is_train_job train job must have set state initializer #380

Closed WonderingWJ closed 2 years ago

WonderingWJ commented 2 years ago

运行deepFM的时候出现错误 F20220826 09:06:46.414013 2273878 replace_embedding_ops_pass.cpp:1160] Check failed: !is_train_job train job must have set state initializer 机器环境 A100x8 运行步骤

  1. 运行docker oneflowinc/oneflow:nightly-cuda11.2
  2. bash train_deepfm.sh
liujuncheng commented 2 years ago

非常感谢您的反馈，我们已经复现问题并将在PR https://github.com/Oneflow-Inc/oneflow/pull/9011 中修复，后续我们将向您提供包括修复的安装包，再次感谢您的反馈

liujuncheng commented 2 years ago

您好，您可以通过下面命令安装更早的版本使用 one embedding

pip install --pre oneflow -f https://staging.oneflow.info/branch/master/cu112/381fb6c6d4536917dd45fd1f48db12b0514732be

我们后续会提供基于修复版本的安装包，再次感谢您的反馈

shangguanshiyuan commented 2 years ago

你好，感谢您的反馈，该问题已修复。 修复该问题的pr已合并至master分支，您更新oneflow的nightly版本或者重新拉取nightly的镜像即可。 再次感谢您的反馈。

python3 -m pip install --pre oneflow -f https://staging.oneflow.info/branch/master/cu112 -U