feat: append attention kernels for fp8 kv-cache #420
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
comaniac
reviewed
Aug 5, 2024
CMakeLists.txt
Outdated
| @@ -91,6 +91,7 @@ set (IDTYPES "i32") | |||
| if(FLASHINFER_ENABLE_FP8) | |||
| list(APPEND DECODE_DTYPES "e4m3" "e5m2") | |||
| list(APPEND DECODE_FP8_DTYPES "e4m3" "e5m2") | |||
| list(APPEND PREFILL_FP8_DTYPES "e4m3") | |||
There was a problem hiding this comment.
e5m2 added, however, this argument only affects C++ tests and has nothing to do with python wheels.
|
@Yard1 @comaniac @cassiewilliam The functionality tests have passed. Feel free to try it and report any possible issues. |
yzh119
added a commit
that referenced
this pull request
Aug 6, 2024
The swizzling mode name in #420 is wrong, this PR aligns it with ptx documentation: 32B -> 64B 64B -> 128B
yzh119
added a commit
that referenced
this pull request
Aug 9, 2024
🤖 I have created a release *beep* *boop* --- ## [0.1.4](v0.1.3...v0.1.4) (2024-08-09) ### Features * append attention kernels for fp8 kv-cache ([#420](#420)) ([906c2f5](906c2f5)) * support min_p sampling ([#422](#422)) ([d52f2da](d52f2da)) * deterministic sampling ([#417](#417)) ([0dd801d](0dd801d)) * more sampling operator options ([#431](#431)) ([68df9c4](68df9c4)) * support fused add rmsnorm ([#419](#419)) ([b781513](b781513)) * support fused silu mul ([#427](#427)) ([ea0ba9a](ea0ba9a)) ### Bug Fixes * fix dispatch fp16 type when enable fp8 ([#430](#430)) ([daa5566](daa5566)) * improve numerical stability of sampling kernels ([#429](#429)) ([898d8ea](898d8ea)) ### Other improvements * break up `_kernels` into multiple modules ([#428](#428)) ([8e482d9](8e482d9)) ### Acknowledgement We thank contributions and feedbacks from the community: [@comaniac](https://github.com/comaniac), [@esmeetu](https://github.com/esmeetu), [@LiuXiaoxuanPKU](https://github.com/LiuXiaoxuanPKU), [@peng1999](https://github.com/peng1999), [@xslingcn](https://github.com/xslingcn), [@Yard1](https://github.com/Yard1), [@zhyncs](https://github.com/zhyncs). --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Zihao Ye <expye@outlook.com>
yzh119
added a commit
that referenced
this pull request
Aug 11, 2024
hardware fp8->fp16 fast conversion instruction is not available for sm_80 & sm_89, which makes #420 slow for these architectures. this pr uses marlin's fast fp8->fp16x4 conversion algorithm (copied from vllm project) to accelerate such cases. Co-authored-by: Antoni Baum <antoni@anyscale.com> Co-authored-by: Cody Yu <cody@anyscale.com>
zhyncs
pushed a commit
that referenced
this pull request
Aug 14, 2024
hardware fp8->fp16 fast conversion instruction is not available for sm_80 & sm_89, which makes #420 slow for these architectures. this pr uses marlin's fast fp8->fp16x4 conversion algorithm (copied from vllm project) to accelerate such cases. Co-authored-by: Antoni Baum <antoni@anyscale.com> Co-authored-by: Cody Yu <cody@anyscale.com>
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.
This implementation do not rely on fp8 tensor cores, but uses fp16 tensor cores instead (so sm_80 architectures can also use it), the fp8 kv-cache will be dequantized on-the-fly.
sm_89 and sm_90 append attention kernels that uses native fp8 tensor cores will be available in later PRs.