2:4 activation sparsity packing kernels #2012
Merged
Conversation
This file contains hidden or 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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2012
Note: Links to docs will display an error until the docs builds have been completed. ✅ You can merge normally! (1 Unrelated Failure)As of commit 126166f with merge base 66eb801 ( BROKEN TRUNK - The following job failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
9 tasks
jcaip
changed the title
jcaip
added
sparsity
topic: new feature
jerryzh168
approved these changes
Apr 29, 2025
liangel-02
pushed a commit
that referenced
this pull request
Aug 25, 2025
This PR is meant to give users the ability to accelerate LLMs with 2:4 activation sparsity, using the approach outlined in our ICLR workshop paper: https://arxiv.org/abs/2503.16672 The main contribution is a cutlass 24_fp8_pack kernel that is able to relatively efficiently calculate the packed data and metadata given a normal dense tensor, which I've copied over from xFormers. ### Performance Benchmarks ``` python benchmarks/benchmark_e2e_fp8_sparse_linear.py | num_tokens | bf16_latency (us) | bf16_c_latency (us) | fp8_c_time (us) | fp8_c_sparse_time (us) | fp8_c_activation_sparse_time (us) | speedup | |-------------:|--------------------:|----------------------:|------------------:|-------------------------:|------------------------------------:|----------:| | 64 | 166.816 | 163.04 | 103.008 | 74.304 | 102.816 | 1.00187 | | 128 | 156.256 | 151.52 | 99.936 | 75.456 | 102.048 | 0.979304 | | 256 | 172.288 | 159.584 | 114.08 | 82.432 | 111.072 | 1.02708 | | 512 | 218.88 | 204.608 | 144.096 | 114.56 | 139.488 | 1.03304 | | 1024 | 394.4 | 392.544 | 251.104 | 196.416 | 227.904 | 1.1018 | | 2048 | 764.608 | 734.816 | 480.704 | 381.152 | 426.688 | 1.12659 | | 4096 | 1658.82 | 1623.58 | 901.344 | 779.008 | 843.392 | 1.06871 | ``` ### Tests ``` pytest tests/sparsity/test_activation24.py ```
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 PR is meant to give users the ability to accelerate LLMs with 2:4 activation sparsity, using the approach outlined in our ICLR workshop paper: https://arxiv.org/abs/2503.16672
The main contribution is a cutlass 24_fp8_pack kernel that is able to relatively efficiently calculate the packed data and metadata given a normal dense tensor, which I've copied over from xFormers.
Performance Benchmarks
Tests