Breaking change: perf: Enable scheduling overlap by default - #4174
Conversation
|
/bot run --disable-fail-fast |
There was a problem hiding this comment.
Pull Request Overview
This PR reverses the overlap scheduler flag by replacing the old “enable_overlap_scheduler” parameter with “disable_overlap_scheduler” (defaulted to False) across the codebase to enable scheduling overlap by default.
- Inverted the flag logic in all relevant modules (worker, executor, pyexecutor, model_engine, decoder, etc.).
- Updated CLI examples, documentation, and configuration files to reflect the new “disable_overlap_scheduler” naming and semantics.
Reviewed Changes
Copilot reviewed 47 out of 47 changed files in this pull request and generated no comments.
Show a summary per file
| File | Description |
|---|---|
| tensorrt_llm/scaffolding/worker.py | Parameter renamed and updated in worker initialization. |
| tensorrt_llm/executor/worker.py | Flag condition inverted for determining scheduling overlap. |
| tensorrt_llm/commands/* | Removed legacy flag usage in serve and eval commands. |
| tensorrt_llm/_torch/pyexecutor/* | Multiple files updated to invert flag logic consistently. |
| examples/* and docs/* | CLI/example scripts and documentation now reference the new flag. |
|
PR_Github #4651 [ run ] triggered by Bot |
|
PR_Github #4651 [ run ] completed with state |
8e4b83f to
dfd073f
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #4692 [ run ] triggered by Bot |
|
PR_Github #4692 [ run ] completed with state |
dfd073f to
3c8cf62
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #4767 [ run ] triggered by Bot |
|
PR_Github #4767 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #4790 [ run ] triggered by Bot |
|
PR_Github #4790 [ run ] completed with state |
|
/bot run --disable-fail-fast --add-multi-gpu-test |
|
PR_Github #5075 [ run ] triggered by Bot |
|
PR_Github #5075 [ run ] completed with state |
|
/bot run --stage-list "DGX_H100-4_GPUs-PyTorch-Others-1" |
|
PR_Github #5127 [ run ] triggered by Bot |
|
PR_Github #5127 [ run ] completed with state |
fdeeb26 to
cc02fae
Compare
|
/bot run --disable-fail-fast --add-multi-gpu-test |
|
PR_Github #5171 [ run ] triggered by Bot |
|
PR_Github #5171 [ run ] completed with state |
Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com> Fix Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com> Fix test_ptp_quickstart_advanced_eagle3 Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com> Fix Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com> Fix Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com> Fix Signed-off-by: Kaiyu Xie <26294424+kaiyux@users.noreply.github.com>
cc02fae to
a0ac332
Compare
|
/bot run --disable-fail-fast --add-multi-gpu-test |
|
PR_Github #5227 [ run ] triggered by Bot |
|
Pre-merge pipeline has passed, merging this in case there are going to be more conflicts. |
|
/bot skip --comment "pre-merge pipeline has passed" |
|
PR_Github #5278 [ skip ] triggered by Bot |
|
PR_Github #5227 [ run ] completed with state |
|
PR_Github #5278 [ skip ] completed with state |
This PR:
enable_overlap_schedulerargument inPyTorchConfigtodisable_overlap_scheduler.