1. 25 Apr, 2022 1 commit
  2. 08 Apr, 2022 1 commit
  3. 01 Apr, 2022 1 commit
  4. 28 Mar, 2022 2 commits
  5. 25 Mar, 2022 1 commit
  6. 16 Mar, 2022 1 commit
  7. 15 Feb, 2022 1 commit
  8. 14 Feb, 2022 1 commit
  9. 26 Jan, 2022 1 commit
  10. 18 Jan, 2022 1 commit
  11. 29 Dec, 2021 1 commit
  12. 17 Nov, 2021 2 commits
  13. 16 Nov, 2021 1 commit
  14. 10 Nov, 2021 1 commit
  15. 02 Nov, 2021 1 commit
  16. 26 Oct, 2021 2 commits
  17. 15 Oct, 2021 1 commit
  18. 12 Oct, 2021 1 commit
  19. 24 Sep, 2021 2 commits
  20. 20 Sep, 2021 1 commit
  21. 10 Sep, 2021 1 commit
  22. 27 Aug, 2021 1 commit
  23. 23 Aug, 2021 1 commit
  24. 12 Aug, 2021 1 commit
  25. 11 Aug, 2021 1 commit
  26. 06 Aug, 2021 1 commit
  27. 05 Aug, 2021 1 commit
  28. 03 Aug, 2021 1 commit
  29. 02 Aug, 2021 1 commit
  30. 01 Aug, 2021 1 commit
  31. 24 Jul, 2021 1 commit
  32. 09 Jul, 2021 1 commit
    • Reed Wanderman-Milne's avatar
      With float16, always use LossScaleOptimizer. · 21286f77
      Reed Wanderman-Milne authored
      Before, it was too easy to accidentally forget to set runtime.loss_scale, which had to always be done if mixed precision is used, otherwise the model would converge to worse accuracy. Now, all that needs to be done to use mixed precision is to set runtime.mixed_precision_dtype=float16.
      
      PiperOrigin-RevId: 383767033
      21286f77
  33. 05 Jul, 2021 1 commit
  34. 01 Jul, 2021 1 commit
  35. 30 Jun, 2021 1 commit
  36. 24 Jun, 2021 1 commit