feat: check stability of output #163
Merged
+3
−358
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Outdated suggestions cannot be applied.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.
fix #170
Sometimes we have bugs on masks, or race condition issues.
They may be catched by checking if the output of the kernel change in time or not.
This is inspired from https://github.com/HazyResearch/flash-attention/blob/main/tests/test_flash_attn.py#L978
The loop is smaller to not make tests too slow
Moreover, old implementations are removed as our kernel is now quite different, old implementations are of little value and make tests slower.