Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[GPU] activations scaling to resolve accuracy issues for infer precision of f16 #27265

Open
wants to merge 40 commits into
base: master
Choose a base branch
from
Open

[GPU] activations scaling to resolve accuracy issues for infer precision of f16 #27265

wants to merge 40 commits into from
+866 −27

Commits on Dec 3, 2024

  1. added the static scaling feature

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    abf9c98 View commit details
    Browse the repository at this point in the history

  2. added a new rt_info scale_factor

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    c8d05b4 View commit details
    Browse the repository at this point in the history

  3. fp16 scaling for vae decoder of sdxl

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    2b2b658 View commit details
    Browse the repository at this point in the history

  4. resolved accuracy issue in transformer of flux.1

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    230c4f2 View commit details
    Browse the repository at this point in the history

  5. removed unnecessary codes

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    eb215f3 View commit details
    Browse the repository at this point in the history

  6. removed unnecessary codes

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    8ec7b80 View commit details
    Browse the repository at this point in the history

  7. renamed to ActivationsScaling

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    9b2d48f View commit details
    Browse the repository at this point in the history

  8. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    f24dce6 View commit details
    Browse the repository at this point in the history

  9. updated to use multiple MatcherPass

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    8897e19 View commit details
    Browse the repository at this point in the history

  10. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    e019834 View commit details
    Browse the repository at this point in the history

  11. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    243e954 View commit details
    Browse the repository at this point in the history

  12. added unit tests

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    267c74e View commit details
    Browse the repository at this point in the history

  13. update code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    76e1aca View commit details
    Browse the repository at this point in the history

  14. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    2656b57 View commit details
    Browse the repository at this point in the history

  15. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    e831e72 View commit details
    Browse the repository at this point in the history

  16. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    54e342c View commit details
    Browse the repository at this point in the history

  17. updated for transformer of FLUX.1

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    eb00539 View commit details
    Browse the repository at this point in the history

  18. disabled FullyConnectedPerLayerScaling

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    4fa0650 View commit details
    Browse the repository at this point in the history

  19. added unit tests

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    409b236 View commit details
    Browse the repository at this point in the history

  20. fixed code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    8a80b6a View commit details
    Browse the repository at this point in the history

  21. Enable FullyConnectedHorizontalFusion with activations scaling

    @andrew-k-park @e-ddykim
    andrew-k-park authored and e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    82c35b6 View commit details
    Browse the repository at this point in the history

  22. updated ScaleDownMultipleLayers

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    736a738 View commit details
    Browse the repository at this point in the history

  23. updated code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    a8f7ca4 View commit details
    Browse the repository at this point in the history

  24. reading ACTIVATIONS_SCALE_FACTOR from rt_info

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    753e870 View commit details
    Browse the repository at this point in the history

  25. updated to use LPT

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    8811c65 View commit details
    Browse the repository at this point in the history

  26. fixed for flux.1 dynamic model

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    4704692 View commit details
    Browse the repository at this point in the history

  27. fix merging faults

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    126fd97 View commit details
    Browse the repository at this point in the history

  28. fixes for flux.1

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    4469c2c View commit details
    Browse the repository at this point in the history

  29. update not to add redundant Convert

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    98cc716 View commit details
    Browse the repository at this point in the history

  30. updated apply_rt_info

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    86c5006 View commit details
    Browse the repository at this point in the history

  31. added a new ScaleDownFusion pass

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    ec9c175 View commit details
    Browse the repository at this point in the history

  32. added a new param useDefaultTransformation for activations scaling

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    300ec4e View commit details
    Browse the repository at this point in the history

  33. update code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    ff0adb0 View commit details
    Browse the repository at this point in the history

  34. update code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    8452092 View commit details
    Browse the repository at this point in the history

  35. updated clamp_fp16 tests

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    a736d9a View commit details
    Browse the repository at this point in the history

  36. code cleanup

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    22fdf6b View commit details
    Browse the repository at this point in the history

  37. code cleanup

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    7683365 View commit details
    Browse the repository at this point in the history

  38. update code style

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    8b966bf View commit details
    Browse the repository at this point in the history

  39. remove redundant code

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    2f77bf8 View commit details
    Browse the repository at this point in the history

  40. updated activations scaling tests

    @e-ddykim
    e-ddykim committed Dec 3, 2024
    Configuration menu
    Copy the full SHA
    f7262b4 View commit details
    Browse the repository at this point in the history