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

Fix hybrid quantization reg issue #27687

Merged
merged 8 commits into from
Nov 29, 2024
Merged

Fix hybrid quantization reg issue #27687

merged 8 commits into from
Nov 29, 2024

Commits on Nov 5, 2024

  1. [GPU] Fixes for hybrid quantization (openvinotoolkit#27127) 

    ### Details:
- set LPT callbacks to handle compression and avoid constant folding for
it (taken from openvinotoolkit#20973)
 - Allow u8/i8 output data type for compressed onednn FC
- Disable Dequantize propagation through Transpose if it's a dependency
of SDPA to keep Transpose+SDPA fusion
    @vladimir-paramuzov @hyunback
    vladimir-paramuzov authored and hyunback committed Nov 5, 2024
    Configuration menu
    Copy the full SHA
    2d3c917 View commit details
    Browse the repository at this point in the history

  2. [GPU] Fix hybrid quantization regression issue. 

    Many daily int8 models have Perf regression.(incorrect conv data type and bias)
Fix kernel selection issue.

Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 5, 2024
    Configuration menu
    Copy the full SHA
    6ed5c74 View commit details
    Browse the repository at this point in the history

Commits on Nov 11, 2024

  1. Fix int8 daily regression models issue. 

    Convolution is expected int8 data type in int8 model, but with mixed weight compressed occurs, it run fp16.

Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 11, 2024
    Configuration menu
    Copy the full SHA
    2e62681 View commit details
    Browse the repository at this point in the history

Commits on Nov 12, 2024

  1. Disable quantized int8 onednn convolution in dynamic. 

    Currently dynamic quantized int8 onednn convolution has the problem. Working with ref convolution.
So replace to run fp16 mode instead.

Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 12, 2024
    Configuration menu
    Copy the full SHA
    f6448ff View commit details
    Browse the repository at this point in the history

  2. Fix the failure when int4 weight compression pattern in lama3.2 

    Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 12, 2024
    Configuration menu
    Copy the full SHA
    0ec2a87 View commit details
    Browse the repository at this point in the history

Commits on Nov 13, 2024

  1. Fix CI func_test failure. 

    Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 13, 2024
    Configuration menu
    Copy the full SHA
    0fc98bb View commit details
    Browse the repository at this point in the history

  2. Back to rnn_joint reg status. 

    Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 13, 2024
    Configuration menu
    Copy the full SHA
    a03dc6a View commit details
    Browse the repository at this point in the history

Commits on Nov 17, 2024

  1. Apply the code review comments. 

    Signed-off-by: hyunback <[email protected]>
    @hyunback
    hyunback committed Nov 17, 2024
    Configuration menu
    Copy the full SHA
    49ad601 View commit details
    Browse the repository at this point in the history