[PT2][Inductor][Reliablity] Fix PT2 latency regression on nanogpt #137248
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137248
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 4141873 with merge base c83178d ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This pull request was exported from Phabricator. Differential Revision: D63806598 |
mengluy0125
changed the title
mengluy0125
force-pushed
the
export-D63806598
branch
from
ce96abb to
67044af
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63806598 |
…torch#137248) Summary: the runtime numeric check in the post grad D63438718 caused PT2 latency increase. We can find all regressions: https://fburl.com/fbdetect/2reizsu9 A failed example: https://www.internalfb.com/servicelab/experiment/4500528673/trial/4500826666 Test Plan: ``` buck2 run mode/opt //caffe2/benchmarks/dynamo/fb:torchbench_run_nanogpt_training -- -m nanogpt -t training ``` defaults-abs_latency-x1000: 51622 ``` servicelab create benchmark_torchbench_run_nanogpt_training -d D63806598 ``` https://www.internalfb.com/servicelab/experiment/4300563668/trial/4300883997 Differential Revision: D63806598
mengluy0125
force-pushed
the
export-D63806598
branch
from
67044af to
4141873
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63806598 |
|
Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as |
2 participants
Differential Revision: D63806598
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire @chauhang @aakhundov