update custom PA kernel with support for fp8 kv cache dtype #87
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
…ustom PA partition size to 512 to prefer throughput scenarios at cost of latency
HaiShaw
reviewed
Jul 18, 2024
| @@ -20,6 +22,20 @@ typedef float16x4 _Half4; | |||
| typedef struct _Half8 { | |||
| _Half4 xy[2]; | |||
| } _Half8; | |||
|
|
|||
There was a problem hiding this comment.
below types defined to unify float16 and bfloat16?
| @@ -15,7 +15,7 @@ | |||
|
|
|||
| # Should be the same as PARTITION_SIZE in `paged_attention_v2_launcher`. | |||
| _PARTITION_SIZE_V1V2 = 512 | |||
| _PARTITION_SIZE_CUSTOM = 256 | |||
| _PARTITION_SIZE_CUSTOM = 512 | |||
There was a problem hiding this comment.
Do we need to keep this different def now?
shajrawi
approved these changes
Sep 12, 2024
|
Have updated this on top of latest main and also enabled FP8 KV cache on BF16. |
|
@mawong-amd I'll take a quick look. |
kkHuang-amd
pushed a commit
that referenced
this pull request
Sep 12, 2024
* update custom PA kernel with support for fp8 kv cache dtype; change custom PA partition size to 512 to prefer throughput scenarios at cost of latency * Fix lint * Fix BF16 with FP8 KV cache (scaled conversion incorrectly done in fp16) --------- Co-authored-by: Matthew Wong <[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Outdated suggestions cannot be applied.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.
custom PA kernel support added for fp8 kv cache dtype.
Change custom PA partition size to 512 to prefer throughput scenarios at cost of latency.
Unit Tested for f16 and fp8 kv cache, num_qheads=64 num_kvheads=8 only.
e2e tested on mlperf offline benchmark.
Unit tests to be updated in separate PR.