-
Notifications
You must be signed in to change notification settings - Fork 2.9k
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
[AutoParallel][PIR] Fit pir grad merge #8985
Conversation
Thanks for your contribution! |
…nto fit_pir_grad_merge
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #8985 +/- ##
===========================================
+ Coverage 53.49% 53.83% +0.33%
===========================================
Files 655 652 -3
Lines 105463 104537 -926
===========================================
- Hits 56419 56275 -144
+ Misses 49044 48262 -782 ☔ View full report in Codecov by Sentry. |
…nto fit_pir_grad_merge
目前 paddlenlp 的 grad_merge 和动转静和动态图 精度对不上,本 pr 暂时先不加单测,后续精度能对齐后补单测 |
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.
LGTM
* fit pir llm grad merge * fit code style * add grad merge llama test * Update run_pretrain_auto.py * Update ci_case_auto.sh * Update ci_case_auto.sh * Update ci_case_auto.sh * Update ci_case_auto.sh * Update ci_case_auto.sh * Update ci_case_auto.sh * Update ci_case_auto.sh * Update ci_case_auto.sh * remove test * Update ci_case_auto.sh * Update ci_case_auto.sh
PR types
Bug fixes
PR changes
Others
Description
现在 pir 的 grad merge 需要和开了 pp 一样,一次性喂进去一个 local_batch 的数据,原来 paddlenlp 在 program ir 下是给数据按照 batch 纬度分开了一次喂一个 micro batch 的数据,这样在 pir 下就会导致纬度不匹配