-
Notifications
You must be signed in to change notification settings - Fork 169
Tuning Utility update #1102
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.
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
Tuning Utility update #1102
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Pull Request Overview
This PR adds environment variable support for tuning utility configuration in the AITER library. It introduces two new environment variables to control logging behavior and allow users to specify custom configuration files for GEMM A4W4 operations.
- Adds
AITER_LOG_TUNED_CONFIGenvironment variable to control logging of tuned shape configurations - Adds
AITER_CONFIG_GEMM_A4W4environment variable to specify custom configuration files for GEMM A4W4 operations - Updates code organization and imports in several files to support these new features
Reviewed Changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| csrc/ck_gemm_a4w4_blockscale/gen_instances.py | Reorganizes imports and updates default tune file path to use the new environment variable |
| csrc/ck_gemm_a4w4_blockscale/gemm_a4w4_blockscale_tune.py | Reorganizes imports and updates default tune file path |
| aiter/ops/gemm_op_a4w4.py | Adds environment variable definitions and conditional logging based on AITER_LOG_TUNED_CONFIG |
| aiter/jit/optCompilerConfig.json | Updates blob generation command to use the new environment variable |
| aiter/jit/core.py | Adds AITER_LOG_TUNED_CONFIG definition and improves type hints |
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
|
Jenkins CI skipped: Check lint failed. Exiting the entire job... |
|
Jenkins CI skipped: Check lint failed. Exiting the entire job... |
* update * fix * fix2 * update * update gemm/fmoe tune * modify README.md * update gen_instances.py * fix lint error * fix typo * rm env AITER_CONFIG_** in ck gen_instances.py * update a8w8_blockscale_bpreshuffle * support to read config from multiple tuned files * fix error in batch_gemm --------- Co-authored-by: Ying.Zhou2 <Ying.Zhou2@amd.com> Co-authored-by: Xin Huang <Xin.Huang@amd.com>
* add asm bf16_gemm to gemm tuner * mdf_fileName (#1160) * Tuning Utility update (#1102) * update * fix * fix2 * update * update gemm/fmoe tune * modify README.md * update gen_instances.py * fix lint error * fix typo * rm env AITER_CONFIG_** in ck gen_instances.py * update a8w8_blockscale_bpreshuffle * support to read config from multiple tuned files * fix error in batch_gemm --------- Co-authored-by: Ying.Zhou2 <Ying.Zhou2@amd.com> Co-authored-by: Xin Huang <Xin.Huang@amd.com> * [MI35X] Fix meta mha 950 error to make guard return bool func itself (#1170) * Make bool guard return func itself * make all return func * write querysol fake * Migrate tests to mi325-1gpu runner due to the slow network issues on TW cluster (#1166) * Temporarily using mi355 runners util issues on mi325 runners are fixed (#1176) * Update sglang_downstream.yaml to use aiter-mi325-1gpu runner (#1181) * CI: Add Aiter Release Package CI (#1171) * update gemmTuner * update tuned_gemm.py * fix lint error * update asm maxSplitK and tuned_gemm.csv * Add AITER_CONFIG_GEMM_BF16 and profile_file params to save all temp results * fix error * add padded_m logic and update gemm_common * refactor get tuned config in tuned_gemm rm m < 256 limitation in gemm_op_a4w4.py * fix lint err * wrapper the get_gemm_a16w16_config using torch_compile_guard * fix lint error * refactor tuned_gemm.py * wrapper gemm_a16w16 with torch_compile_guard * fix lint error * add gen_fake func * update * updae GemmTuner --profile_file --------- Co-authored-by: amd-ruitang3 <145657428+amd-ruitang3@users.noreply.github.com> Co-authored-by: Lingpeng Jin <103567126+valarLip@users.noreply.github.com> Co-authored-by: Xin Huang <Xin.Huang@amd.com> Co-authored-by: ZhangLirong <lirzhang@amd.com> Co-authored-by: valarLip <340077269@qq.com>
* update * fix * fix2 * update * update gemm/fmoe tune * modify README.md * update gen_instances.py * fix lint error * fix typo * rm env AITER_CONFIG_** in ck gen_instances.py * update a8w8_blockscale_bpreshuffle * support to read config from multiple tuned files * fix error in batch_gemm --------- Co-authored-by: Ying.Zhou2 <Ying.Zhou2@amd.com> Co-authored-by: Xin Huang <Xin.Huang@amd.com>
* add asm bf16_gemm to gemm tuner * mdf_fileName (ROCm#1160) * Tuning Utility update (ROCm#1102) * update * fix * fix2 * update * update gemm/fmoe tune * modify README.md * update gen_instances.py * fix lint error * fix typo * rm env AITER_CONFIG_** in ck gen_instances.py * update a8w8_blockscale_bpreshuffle * support to read config from multiple tuned files * fix error in batch_gemm --------- Co-authored-by: Ying.Zhou2 <Ying.Zhou2@amd.com> Co-authored-by: Xin Huang <Xin.Huang@amd.com> * [MI35X] Fix meta mha 950 error to make guard return bool func itself (ROCm#1170) * Make bool guard return func itself * make all return func * write querysol fake * Migrate tests to mi325-1gpu runner due to the slow network issues on TW cluster (ROCm#1166) * Temporarily using mi355 runners util issues on mi325 runners are fixed (ROCm#1176) * Update sglang_downstream.yaml to use aiter-mi325-1gpu runner (ROCm#1181) * CI: Add Aiter Release Package CI (ROCm#1171) * update gemmTuner * update tuned_gemm.py * fix lint error * update asm maxSplitK and tuned_gemm.csv * Add AITER_CONFIG_GEMM_BF16 and profile_file params to save all temp results * fix error * add padded_m logic and update gemm_common * refactor get tuned config in tuned_gemm rm m < 256 limitation in gemm_op_a4w4.py * fix lint err * wrapper the get_gemm_a16w16_config using torch_compile_guard * fix lint error * refactor tuned_gemm.py * wrapper gemm_a16w16 with torch_compile_guard * fix lint error * add gen_fake func * update * updae GemmTuner --profile_file --------- Co-authored-by: amd-ruitang3 <145657428+amd-ruitang3@users.noreply.github.com> Co-authored-by: Lingpeng Jin <103567126+valarLip@users.noreply.github.com> Co-authored-by: Xin Huang <Xin.Huang@amd.com> Co-authored-by: ZhangLirong <lirzhang@amd.com> Co-authored-by: valarLip <340077269@qq.com>
Motivation
Technical Details
Test Plan
Test Result
Submission Checklist