Load dynamic module (remote code) only once if code isn't change (#33162)

* Load remote code only once

* Use hash as load indicator

* Add a new option `force_reload` for old behavior (i.e. always reload)

* Add test for dynamic module is cached

* Add more type annotations to improve code readability

* Address comments from code review
This commit is contained in:
Xuehai Pan
2024-09-06 19:49:35 +08:00
committed by GitHub
parent 1bd9d1c899
commit e1c2b69c34
6 changed files with 139 additions and 12 deletions

View File

@@ -116,12 +116,29 @@ class AutoFeatureExtractorTest(unittest.TestCase):
)
self.assertEqual(feature_extractor.__class__.__name__, "NewFeatureExtractor")
# Test the dynamic module is loaded only once.
reloaded_feature_extractor = AutoFeatureExtractor.from_pretrained(
"hf-internal-testing/test_dynamic_feature_extractor", trust_remote_code=True
)
self.assertIs(feature_extractor.__class__, reloaded_feature_extractor.__class__)
# Test feature extractor can be reloaded.
with tempfile.TemporaryDirectory() as tmp_dir:
feature_extractor.save_pretrained(tmp_dir)
reloaded_feature_extractor = AutoFeatureExtractor.from_pretrained(tmp_dir, trust_remote_code=True)
self.assertEqual(reloaded_feature_extractor.__class__.__name__, "NewFeatureExtractor")
# The feature extractor file is cached in the snapshot directory. So the module file is not changed after dumping
# to a temp dir. Because the revision of the module file is not changed.
# Test the dynamic module is loaded only once if the module file is not changed.
self.assertIs(feature_extractor.__class__, reloaded_feature_extractor.__class__)
# Test the dynamic module is reloaded if we force it.
reloaded_feature_extractor = AutoFeatureExtractor.from_pretrained(
"hf-internal-testing/test_dynamic_feature_extractor", trust_remote_code=True, force_download=True
)
self.assertIsNot(feature_extractor.__class__, reloaded_feature_extractor.__class__)
def test_new_feature_extractor_registration(self):
try:
AutoConfig.register("custom", CustomConfig)