Avoid converting k and v to q dtype #2201
Closed
Conversation
This file contains hidden or 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
Summary: We don't want to hard code k and v into q's dtype, instead we should make sure they are always the same before feeding into sdpa. The problem was due to dtype mismatch between the kv cache used for tracing and the actual weights. This happens in the fp16 flow. After we convert the whole model to fp16, we still use fp32 kv cache tensors for tracing and that's causes the dtype mismatch issue. This diff changes the logic to be using the same dtype as the weights, for kv cache during tracing Differential Revision: D54426672
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/2201
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit a768904 with merge base c3c667e ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D54426672 |
mikekgfb
approved these changes
Mar 1, 2024
|
This pull request has been merged in 5a18cc6. |
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Outdated suggestions cannot be applied.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.
Summary:
We don't want to hard code k and v into q's dtype, instead we should make sure they are always the same before feeding into sdpa.
The problem was due to dtype mismatch between the kv cache used for tracing and the actual weights. This happens in the fp16 flow. After we convert the whole model to fp16, we still use fp32 kv cache tensors for tracing and that's causes the dtype mismatch issue.
This diff changes the logic to be using the same dtype as the weights, for kv cache during tracing
Differential Revision: D54426672