feat: support bmm fp8 #469
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
yzh119
reviewed
Aug 26, 2024
|
Another suggestion is to move group gemm and bmm fp8 to a common gemm.py, we should also update the group_gemm.rst (to gemm.rst) as well. |
make sense |
yzh119
reviewed
Aug 26, 2024
yzh119
added a commit
that referenced
this pull request
Aug 27, 2024
The documentation was not indexed properly in #469 , this PR fixes the issue.
yzh119
added a commit
that referenced
this pull request
Aug 27, 2024
🤖 I have created a release *beep* *boop* --- ## [0.1.6](v0.1.5...v0.1.6) (2024-08-27) ### SM75 Support Starting from [0.1.6](v0.1.5...v0.1.6), our pre-built wheels include experimental support sm75 (Turing architecture GPUs such as Tesla T4, Quadro RTX 6000 and RTX 2080). ### API Changes #### `plan`/`run` Since [0.1.6](v0.1.5...v0.1.6) on, `begin_forward`/`forward`/`end_forward` APIs are replaced with the new `plan`/`run` API. - `forward` is renamed to `run`, which is more precise and consistent with the naming convention of cutlass's python API. - `begin_forward` is renamed to `plan`, which is consistent with the naming convention of nvmath API. - `end_forward` is deprecated and has no effect after this PR. There is some slight difference between the old `forward` and the new `run` API: - All extra arguments such as `causal` and `logits_soft_cap` will be provided in `plan` (previously `begin_forward`) API, and cached until next `plan` call, and we only need to provide query and KV-Cache tensors in `run` API. The old `begin_forward`/`forward`/`end_forward` APIs are still functional, but we will gradually deprecate them in future releases. Check [#466](#466) for more details. #### `MultiLevelCascadeAttentionWrapper` Since [0.1.6](v0.1.5...v0.1.6) on, we introduce a new `MultiLevelCascadeAttentionWrapper` API for cascade inference, which supports multi-level cascade inference where all levels' KV-Cache can be managed in a unified Paged KV-Cache. See [documentation](https://docs.flashinfer.ai/api/python/cascade.html#flashinfer.cascade.MultiLevelCascadeAttentionWrapper) and [tutorial](https://docs.flashinfer.ai/tutorials/kv_layout.html#multi-level-cascade-inference-data-layout) on API usage and layout explaination. The old `BatchDecodeWithSharedPrefixPagedKVCacheWrapper` and `BatchPrefillWithSharedPrefixPagedKVCacheWrapper` will be deprecated in future releases. ### Features * sm75 support ([#448](#448), [#449](#449)) * add `MultiLevelCascadeAttentionWrapper` API ([#462](#462)) ([1e37989](1e37989)) * add accept num, emit num metric for ChainSpeculativeSampling ([#450](#450)) ([fa38b5e](fa38b5e)) * support bmm fp8 ([#469](#469)) ([f1c0b68](f1c0b68)) ### Refactor * refactor: replace `begin_forward`/`forward`/`end_forward` with `plan`/`run` [#466](#466) ### Misc * misc: improve error handling of sampling kernels ([#456](#456)) ([0dce178](0dce178)) ### Performance Improvements * slight optimization on f16->f8 fragment layout swizzling ([#453](#453)) ([0d61871](0d61871)) * slight optimization on fragment layout swizzle ([#458](#458)) ([7c397cb](7c397cb)) * use persistent kernel for merging attention states ([#459](#459)) ([be6bf5b](be6bf5b)) ### Acknowledgement We thank [@LiuXiaoxuanPKU](https://github.com/LiuXiaoxuanPKU) on enhance of speculative sampling operator, [@merrymercy](https://github.com/merrymercy) on API change suggestion and [@zhyncs](https://github.com/zhyncs) on integrating fp8 BMM cublas implementation. --- 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>
This was referenced Aug 30, 2024
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.
torch.bmmdoesn't support fp8 andtorch._scaled_mmdoesn't support 3d, so I write this one. @yzh119 cc @merrymercy @Ying1123 @ispobockThanks @yzh119 for assisting with debug.
AType: fp8 e4m3, fp8 e5m2
BType: fp8 e4m3, fp8 e5m2
DType: bf16, fp16
Does not support both AType and BType fp8 e5m2. ref https://docs.nvidia.com/cuda/cublas/#cublasltmatmul
works on H100