Integrate MLX SDPA kernels with mask #2820
Open
+2,149
−963
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
ivarflakstad
approved these changes
Apr 17, 2025
There was a problem hiding this comment.
Great work 🎉
I only have 2 comments really.
One is that at this point it's really time to start precompiling kernels. Not for performance, just for the sake of project structure and maintainability.
I have that ready to go so I'll make a PR soon.
The second is that I wonder if it makes sense to move more of the logic in candle-nn into candle-metal-kernels.
The standard asserts etc is fine, but maybe figuring out the correct kernel to call is actually a concept that belongs inside candle-metal-kernels?
That's open for discussion obviously.
In any case this is ready to merge in my opinion - if all tests pass and it runs smoothly
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Outdated suggestions cannot be applied.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 integrates kernel developments from: ml-explore/mlx#1924.
Specifically, our
candle_nn::ops::sdpafunction now dispatches to optimized implementations for with and without prompts. There is also an option for causal masking, removing the necessity for mask materialization.Overall, this means that we can fuse the attention operation on Metal for prompt and decode phases!
I will update this PR further with benchmarks, but it is tested and working in my fork through mistral.rs.