Apply gradients with distributed_shampoo fails

未关闭
#285 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
3/5
预计耗时
1-2 天
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
停滞
技术栈
python, tensorflow

调研方向

The traceback points to lingvo/core/distributed_shampoo.py, in _create_slots around line 336, where a TensorFlow variable is used as a dictionary key. Start by reading that method and the TensorFlow 2.7 variable hashing behavior described by the error. Confirm the failure with the reported apply_gradients() call; done means distributed shampoo applies gradients successfully with TensorFlow 2.7.1.

由索引模型根据 Issue 内容生成。

描述

Hello,

I am trying to run distributed shampoo implemented in lingvo. I have the compatible tensorflow version with lignvo.

Python 3.8.12 | packaged by conda-forge | (default, Oct 12 2021, 21:59:51)
[GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> tf.__version__
'2.7.1'
>>>

But when I do opt.apply_gradients(), it throws an error about tf.variable being unhashable. apply_gradients() works fine with other optimizers.

File "program.py", line 177, in train_step
  optimizer.apply_gradients(
File "/home/ubuntu/anaconda3/envs/tensorflow2_p38/lib/python3.8/site-packages/tensorflow/python/training/optimizer.py", line 605, in apply_gradients
  self._create_slots(var_list)
File "/home/ubuntu/anaconda3/envs/tensorflow2_p38/lib/python3.8/site-packages/lingvo/core/distributed_shampoo.py", line 336, in _create_slots
  self._partitioner_metadata[v] = TensorPartitioner.partition_metadata(
File "/home/ubuntu/anaconda3/envs/tensorflow2_p38/lib/python3.8/site-packages/tensorflow/python/ops/variables.py", line 1085, in __hash__
  raise TypeError(
TypeError: Variable is unhashable. Instead, use variable.ref() as the key. (Variable: <tf.Variable 'conv2d_23/kernel:0' shape=(7, 7, 8, 32) dtype=float32, numpy=
array([[[[-6.29460454e-01, -4.73986834e-01, -3.37867022e-01, ...,
         5.98385096e-01, -1.81423143e-01,  4.59209830e-02],
       [ 1.96352646e-01, -1.11289904e-01,  3.36863637e-01, ...,

Appreciate any help to fix the issue.

主要语言
Python
星标
2.9k
派生
452
PR 合并指标
30 天内没有已合并 PR

环境准备

  • 没有 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

tensorflow/lingvo 的其他 Issue

查看 tensorflow/lingvo 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。