jit: further accelerate compilation by spliting files and multi-threading #628
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
added a commit
that referenced
this pull request
Nov 24, 2024
Currently unittests are slow when using flashinfer jit because we only compile kernels the first time we run it, it's blocking and didn't compile multiple ops in parallel. This PR add a warmup pre-hook to kernel unittests, so that we compile all necessary kernels before running the unittests in JIT mode, which greatly accelerate the unittests. This PR also fixes the several issues with #628 : 1. using thread-safe `make_dirs(..., exist_ok=True)` instead of relying on `os.path.exists` 2. change the signature of `parallel_load_modules` to lists of `(jit_module_creation_func, args)` instead of lambda function, because lambda function captures variable by ref instead of value, which may cause some unexpected errors.
yzh119
added a commit
that referenced
this pull request
Nov 24, 2024
Followup of #628, this PR splits prefill attention jit templates so that we compile different mask modes in different files. JIT compilation time of a prefill kernels of a certain configuration (shape, dtype etc) could be reduced to 10 seconds after this PR.
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Outdated suggestions cannot be applied.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 accelerates JIT compilation by:
parallel_load_modulesfunction to load necessary modules for a model in parallel using python multi-threading.The batch prefill attention template could be further split into multiple instances to accelerate compilation, we leave that for future work.