Add support for FP8 KV cache scales #2628
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
danieldk
force-pushed
the
feature/fp8-kv-cache-scale
branch
3 times, most recently
from
08c0b3f to
98efcb4
Compare
Since FP8 only has limited dynamic range, we can scale keys/values before storing them into the cache (and unscale them in attention). To avoid rescaling the cache as the absmax values change, good scales are usually determined per layer using calibration calibration data and stored in the checkpoint. This change adds support for for using key-value scales and loading them from checkpoints in the two most common formats: - Separate per-layer `k_scale` and `v_scale` scalars. - Per-layer `kv_scale` scalar (older format). Currently, scales are only used with an `float8_e4m3fn` cache. Besides adding support for key/value scales, the `fp8_quantize` function is also extended to support quantization with a kernel vendored from vLLM. This is slightly faster than the PyTorch implementation, but also scales in FP32, potentially improving accuracy.
danieldk
force-pushed
the
feature/fp8-kv-cache-scale
branch
from
98efcb4 to
1f18cb6
Compare
mht-sharma
reviewed
Oct 24, 2024
There was a problem hiding this comment.
LGTM! Thanks for the PR @danieldk. This will help me enable FP8 KV cache on ROCm next.
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Outdated suggestions cannot be applied.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.
What does this PR do?
Since FP8 only has limited dynamic range, we can scale keys/values before storing them into the cache (and unscale them in attention). To avoid rescaling the cache as the absmax values change, good scales are usually determined per layer using calibration calibration data and stored in the checkpoint.
This change adds support for for using key-value scales and loading them
from checkpoints in the two most common formats:
k_scaleandv_scalescalars.kv_scalescalar (older format).Currently, scales are only used with an
float8_e4m3fncache.Besides adding support for key/value scales, the
fp8_quantizefunction is also extended to support quantization with a kernel vendored from vLLM. This is slightly faster than the PyTorch implementation, but also scales in FP32, potentially improving accuracy.Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.