[Inductor] Further tune block size for templated attention on H100 #125286
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/125286
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 93fb304 with merge base 4d5f807 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Chillee
approved these changes
May 1, 2024
There was a problem hiding this comment.
Should probably check if it exactly matches
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
petrex
pushed a commit
to petrex/pytorch
that referenced
this pull request
May 3, 2024
…ytorch#125286) Run a script to enumerate and get the best default block size for templated attention. A100 -> no change, check numbers at pytorch#125139 H100 ## torch.bfloat16 Before: ``` | Type | Speedup | batch_size | num_heads | q_seq_len | k_seq_len | head_dim | score_mod | dtype | |---------|-----------|--------------|-------------|-------------|-------------|------------|---------------|----------------| | Average | 1.103 | | | | | | | | | Max | 1.322 | 8 | 16 | 512 | 512 | 64 | noop | torch.bfloat16 | | Min | 0.829 | 1 | 16 | 1024 | 1024 | 128 | relative_bias | torch.bfloat16 | ``` After: ``` | Type | Speedup | batch_size | num_heads | q_seq_len | k_seq_len | head_dim | score_mod | dtype | |---------|-----------|--------------|-------------|-------------|-------------|------------|---------------|----------------| | Average | 1.137 | | | | | | | | | Max | 1.442 | 1 | 16 | 512 | 512 | 128 | relative_bias | torch.bfloat16 | | Min | 0.913 | 1 | 16 | 1024 | 1024 | 64 | head_bias | torch.bfloat16 | ``` ## torch.float32 Before: ``` | Type | Speedup | batch_size | num_heads | q_seq_len | k_seq_len | head_dim | score_mod | dtype | |---------|-----------|--------------|-------------|-------------|-------------|------------|---------------|---------------| | Average | 2.269 | | | | | | | | | Max | 3.740 | 16 | 16 | 1024 | 1024 | 64 | noop | torch.float32 | | Min | 0.761 | 1 | 16 | 512 | 512 | 128 | relative_bias | torch.float32 | ``` After: ``` | Type | Speedup | batch_size | num_heads | q_seq_len | k_seq_len | head_dim | score_mod | dtype | |---------|-----------|--------------|-------------|-------------|-------------|------------|-------------|---------------| | Average | 2.489 | | | | | | | | | Max | 3.755 | 16 | 16 | 4096 | 4096 | 64 | noop | torch.float32 | | Min | 1.609 | 1 | 16 | 512 | 512 | 64 | head_bias | torch.float32 | ``` Pull Request resolved: pytorch#125286 Approved by: https://github.com/Chillee
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Run a script to enumerate and get the best default block size for templated attention.
A100 -> no change, check numbers at #125139
H100
torch.bfloat16
Before:
After:
torch.float32
Before:
After:
cc @ezyang @msaroufim @bdhirsh @anijain2305 @chauhang @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @peterbell10 @ipiszy @yf225 @chenyang78 @kadeng @muchulee8 @ColinPeppler @amjames @desertfire