[CPU] Enabled FP16 Compressed FC on models with PagedAttention #26279
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
dmitry-gorokhov
added
category: CPU
dmitry-gorokhov
force-pushed
the
feature/fp16_compressed_fc
branch
from
4bd98c1
to
438403b
Compare
ilya-lavrenov
approved these changes
Oct 12, 2024
usstq
approved these changes
Oct 14, 2024
There was a problem hiding this comment.
LGTM, vcvtph2ps used for decompressing FP16 into FP32 introduced overhead according to tests shown in https://jira.devtools.intel.com/browse/CVS-133453, but given the fact:
- most serving workload with big-batch-size (compute-bounded) would be running on Xeon with (src-bf16, wei-bf16) or (src-f16, wei-f16) kernel
- most likely workload running on AI PC, would use small batch-size and (src-f32, wei-f16) can help no matter it's SDPA or PA.
so we can safely enable it
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Outdated suggestions cannot be applied.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.
No description provided.