Skip to content

GitLab

"test/ut/experiment/test_exp_config.py" did not exist on "cb090e8c9bc832fe5c0e2019831ef83ccd56fb55"
Switch branch/tag
  1. 12 Apr, 2021 1 commit
    • Reed Wanderman-Milne's avatar
      Use nonexperimental mixed precision API for official models. · ba8ad4f5
      Reed Wanderman-Milne authored 
      For all modified calls to set_mixed_precision_policy(), the loss_scale argument was removed, as it cannot be passed if the nonexperimental API is used. For all such callers, the loss_scale is later used to explicitly create a LossScaleOptimizer, so removing the argument has no impact.

Switching to the non-experimental LossScaleOptimizer has no effect, as it has near identical behavior and all isinstance checks within the official models check for the non-experimental version.

PiperOrigin-RevId: 368101975
      ba8ad4f5
  3. 10 Apr, 2021 1 commit
    • Reed Wanderman-Milne's avatar
      Remove dynamic_loss_scale argument to define_performance. · 3803472a
      Reed Wanderman-Milne authored 
      All models which support loss scaling support dynamic loss scaling, so the argument has no purpose. It used to be that some models scaled the loss manually instead of using a LossScaleOptimizer, and so did not support dynamic loss scaling.

PiperOrigin-RevId: 367719521
      3803472a
  5. 08 Apr, 2021 1 commit
  7. 07 Apr, 2021 1 commit
  9. 06 Apr, 2021 2 commits
    • Hongkun Yu's avatar
      Clarify the deprecation warning for bert/ readme. · 882f8259
      Hongkun Yu authored 
      PiperOrigin-RevId: 367101911
      882f8259
    • Jeremiah Liu's avatar
      Disable temperature scaling during training. · ff3ed4cc
      Jeremiah Liu authored 
      For the `GaussianProcessClassificationHead`, the temperature scaling needs to be disabled during training to avoid unexpected modification to the learning rate, which harms model quality. (Unfortunately, this seems to require adding `training` to the `call` method).

Also set the default of `gp_cov_ridge_penalty` in `RandomFeatureGaussianProcess` to 1 to be consistent with that in the `GaussianProcessClassificationHead`.

PiperOrigin-RevId: 366917075
      ff3ed4cc
  11. 05 Apr, 2021 2 commits
  13. 01 Apr, 2021 5 commits
  15. 29 Mar, 2021 1 commit
  17. 26 Mar, 2021 3 commits
  19. 25 Mar, 2021 1 commit
  21. 24 Mar, 2021 1 commit
  23. 23 Mar, 2021 1 commit
  25. 22 Mar, 2021 4 commits
  27. 20 Mar, 2021 2 commits
  29. 18 Mar, 2021 2 commits
  31. 17 Mar, 2021 2 commits
  33. 16 Mar, 2021 6 commits
  35. 14 Mar, 2021 2 commits
  37. 12 Mar, 2021 2 commits