Fix Throughputbenchmark issue #144669
Fix Throughputbenchmark issue #144669
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/144669
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New FailuresAs of commit b5f6e1c with merge base 6e77d7c ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This PR needs a
|
| @@ -539,6 +539,9 @@ struct AutocastState { | |||
|
|
|||
| bool operator==(const AutocastState& o) const { | |||
| for (size_t i = 0; i < DEVICES.size(); i++) { | |||
| if (enabled[i] == false && o.enabled[i] == false) { | |||
| continue; | |||
There was a problem hiding this comment.
Please elaborate why skips the check of dtype when enabled is False.
| for (size_t i = 0; i < DEVICES.size(); i++) { | ||
| at::autocast::set_autocast_enabled(DEVICES[i], autocast_enabled[i]); | ||
| at::autocast::set_autocast_dtype(DEVICES[i], autocast_dtype[i]); | ||
| } |
There was a problem hiding this comment.
Do we need to check the status of cache_enabled?
Fixes 144461
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames