Fix _cache_founf_inf #7997
Fix _cache_founf_inf #7997
Conversation
|
Thanks for your contribution! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #7997 +/- ##
===========================================
+ Coverage 56.42% 56.55% +0.13%
===========================================
Files 589 592 +3
Lines 90252 91040 +788
===========================================
+ Hits 50921 51484 +563
- Misses 39331 39556 +225 ☔ View full report in Codecov by Sentry. |
examples/RLHF/ppo_trainer.py
Outdated
| @@ -415,7 +415,7 @@ def full_training_step(self: Trainer, inputs: Dict[str, paddle.Tensor], **kwargs | |||
| self.scaler.step(self.optimizer) | |||
| self.scaler.update() | |||
| scale_after = self.scaler._scale | |||
| optimizer_was_run = not self.scaler._cache_founf_inf | |||
| optimizer_was_run = not self.scaler._cache_found_inf | |||
There was a problem hiding this comment.
Please check if there's a way to ensure compatibility between interfaces. We've encountered an issue where using PaddleNLP's development version with PaddlePaddle v2.6.0 results in incompatibility.
There was a problem hiding this comment.
Changed to check _cache_founf_inf
|
@wawltor It has been modified. codecov suggested that there was a coverage problem. Do I need to add the test, Where is it to add |
PR types
Others
PR changes
Others
Description
Fix _cache_founf_inf
Paddle也需要修改 PaddlePaddle/Paddle#61954