Fix bug with number of evaluation steps #384
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
we were running way to many evaluation steps if the model is pipe parallel + has g.a.s on because of this line
```python
for _ in range(neox_args.gradient_accumulation_steps):
```
- fixing this to 1 if the model is pipe parallel fixes the issue, as .eval_batch() already takes gradient accumulation steps into account.
sweinbach
approved these changes
Jul 30, 2021
|
Tested in a running training as well... |
sdtblck
added a commit
that referenced
this pull request
Aug 21, 2021
* optimize data preprocessing semaphore is a little too small and slows down tokenizing * Make killall.sh less bruteforce * [temporary] fix to index errors * [temporary] fix to index errors * print sizes of tensors when inspecting checkpoint (#382) Co-authored-by: Samuel Weinbach <[email protected]> * Use lru_cache for GPT2Tokenizer.bpe (#383) GPT2Tokenizer currently uses an unbounded cache, which causes very high memory usage with tools/preprocess_data.py * Fix bug with number of evaluation steps (#384) we were running way to many evaluation steps if the model is pipe parallel + has g.a.s on because of this line ```python for _ in range(neox_args.gradient_accumulation_steps): ``` - fixing this to 1 if the model is pipe parallel fixes the issue, as .eval_batch() already takes gradient accumulation steps into account. * Create CITATION.cff * Update CITATION.cff * Update documentation (#392) * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * add info about installing fused kernels * Update README.md * Update README.md * sparsity + minor typos add the instructions to install triton * change path to ssd-1 * typo * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md * Update README.md Co-authored-by: Shivanshu Purohit <[email protected]> Co-authored-by: Stella Biderman <[email protected]> Co-authored-by: Samuel Weinbach <[email protected]> Co-authored-by: Samuel Weinbach <[email protected]> Co-authored-by: iczero <[email protected]> Co-authored-by: Shivanshu Purohit <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
we were running way too many evaluation steps if the model is pipe parallel + has g.a.s on because of this line
fixing this to 1 if the model is pipe parallel fixes the issue, as .eval_batch() already takes gradient accumulation steps into account.
This also relates to this issue, and is probably the reason we were seeing a stopiteration error