Skip to content

GitLab

Switch branch/tag
  1. 08 Dec, 2021 1 commit
    • Smit Hinsu's avatar
      Replace dynamic Squeeze with reshape to 1d · ea36aff1
      Smit Hinsu authored 
      TPUs don't support dynamic Squeeze as it won't be able to identify the dimensions to remove if the input shape is dynamic. For example, [<=16, 1] could return a vector or scalar result depending on the size of the dynamic dim.

PiperOrigin-RevId: 415083667
      ea36aff1
  3. 01 Dec, 2021 2 commits
  5. 17 Nov, 2021 1 commit
  7. 27 Aug, 2021 2 commits
  9. 06 Jun, 2021 4 commits
  11. 12 Apr, 2021 2 commits
    • 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
    • Reed Wanderman-Milne's avatar
      Use nonexperimental mixed precision API for official models. · 0d8f9807
      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
      0d8f9807
  13. 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
  15. 09 Apr, 2021 1 commit
    • Reed Wanderman-Milne's avatar
      Remove dynamic_loss_scale argument to define_performance. · e353e4e5
      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
      e353e4e5
  17. 06 Apr, 2021 2 commits
  19. 10 Mar, 2021 2 commits
  21. 25 Feb, 2021 4 commits
  23. 05 Jan, 2021 2 commits
  25. 22 Dec, 2020 2 commits
  27. 21 Dec, 2020 1 commit
  29. 19 Dec, 2020 2 commits
    • Hongkun Yu's avatar
      Clean up · efa0c440
      Hongkun Yu authored 
      PiperOrigin-RevId: 348291651
      efa0c440
    • Hongkun Yu's avatar
      Clean up · 10939e58
      Hongkun Yu authored 
      PiperOrigin-RevId: 348291651
      10939e58
  31. 15 Dec, 2020 2 commits
  33. 14 Dec, 2020 4 commits
  35. 25 Nov, 2020 2 commits
  37. 17 Nov, 2020 2 commits
  39. 16 Nov, 2020 1 commit