feat(autotuner): Make autotune cache class configurable via env var #1071
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
meta-cla
bot
added
the
CLA Signed
jansel
requested changes
Nov 4, 2025
Contributor
jansel
left a comment
jansel
left a comment
There was a problem hiding this comment.
Perhaps we should make this more generic (similar to settings.autotuner_fn) so that as we add more classes we the config will still work.
Can you add a test?
fulvius31
force-pushed
the
strict-cache
branch
from
4bbb119 to
18c0b14
Compare
fulvius31
changed the title
Contributor
Author
|
@jansel Thanks for the review! Let me know if it looks good now. |
jansel
approved these changes
Nov 4, 2025
fulvius31
force-pushed
the
strict-cache
branch
from
18c0b14 to
54add4e
Compare
Contributor
Author
|
@jansel I don't know if the cpu fail is caused by this PR. I've added a baseline for the other failed test. |
Contributor
Author
|
@jansel it looks like the cpu workflow does not install triton for some reason that does not depend on this PR |
Adds a new boolean setting, `use_strict_cache`, to `helion.runtime.Settings`, controlled by the `HELION_USE_STRICT_CACHE` environment variable. The `default_autotuner_fn` now reads this setting and selects `StrictLocalAutotuneCache` if set to true, falling back to `LocalAutotuneCache` otherwise. This allows users to easily enable stricter cache validation without needing to write a custom `autotuner_fn`. Signed-off-by: Alessandro Sangiorgi <asangior@redhat.com>
Adds a new `HELION_AUTOTUNE_CACHE` environment variable to allow
users to select the autotuner cache implementation, similar to how
`HELION_AUTOTUNER` selects the search algorithm.
- `helion/autotuner/__init__.py`:
- Adds a `cache_classes` dictionary to register available cache
implementations (LocalAutotuneCache, StrictLocalAutotuneCache).
- `helion/runtime/settings.py`:
- Adds a new string setting, `autotune_cache`, which defaults
to "LocalAutotuneCache" and is controlled by the
`HELION_AUTOTUNE_CACHE` environment variable.
- Updates `default_autotuner_fn` to use this new setting to
look up and instantiate the correct cache class from
`autotuner.cache_classes`.
- `test_autotuner.py`:
- Adds the `TestAutotuneCacheSelection` test case.
- Verifies that the default cache is `LocalAutotuneCache`.
- Verifies that `HELION_AUTOTUNE_CACHE=StrictLocalAutotuneCache`
correctly selects `StrictLocalAutotuneCache`.
- Verifies that an invalid cache name raises a `ValueError`.
Signed-off-by: Alessandro Sangiorgi <asangior@redhat.com>
fulvius31
force-pushed
the
strict-cache
branch
from
e08b536 to
32fc11d
Compare
Contributor
Author
|
@jansel I've fixed the issue with the cpu environment and tested locally |
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.
Adds a new
HELION_AUTOTUNE_CACHEenvironment variable to allowusers to select the autotuner cache implementation, similar to how
HELION_AUTOTUNERselects the search algorithm.helion/autotuner/__init__.py:cache_classesdictionary to register available cacheimplementations (LocalAutotuneCache, StrictLocalAutotuneCache).
helion/runtime/settings.py:autotune_cache, which defaultsto "LocalAutotuneCache" and is controlled by the
HELION_AUTOTUNE_CACHEenvironment variable.default_autotuner_fnto use this new setting tolook up and instantiate the correct cache class from
autotuner.cache_classes.test_autotuner.py:TestAutotuneCacheSelectiontest case.LocalAutotuneCache.HELION_AUTOTUNE_CACHE=StrictLocalAutotuneCachecorrectly selects
StrictLocalAutotuneCache.ValueError.