pip installation issue

[c0derzer0]

I am tying to pip install lingvo==0.9.0. Although I have tensorflow-gpu 2.6, it downloads cpu tensorflow 2.5 which breaks rest of my dependencies. Is there a way to rectify this? Does git clone help in this case or does is it need tensorflow 2.5 to run. Specifically I just want to use distributed shampoo optimizer in the library for my problem. I'd appreciate it if you could point to any standalone implementation of it on tensorflow.

[laurentes]

I think lingvo 0.9.0 was Pip packaged with Requires-Dist: tensorflow (==2.5.0).
Would you be able to use lingvo 0.10.0 instead, which includes Requires-Dist: tensorflow (~=2.6.0)?

[c0derzer0]

Thanks for the suggestion. Looks like having the corresponding tensorflow-gpu version rectified the issue. Additionally I have a question on how to use the distributed shampoo code.
from the distributed_shampoo_test.py is the follow code snippet enough

opt = distributed_shampoo.DistributedShampoo(
      learning_rate=1.0,
      momentum=0.0,
      start_preconditioning_steps=0,
      synchronous_preconditioning=True,
      global_step=global_step)

  # Run a single step of gradient update.
  update = opt.apply_gradients(zip([grad], [var]), global_step=global_step)

I have not looked into the distributed shampoo algorithm deeply yet. Any suggestions would be appreciated