SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fixing tests for Perceiver (#14739)

Browse Source 
* Adding some slow test to check for perceiver at least from a high level.

* Re-enabling fast tests for Perceiver ImageClassification.

* Perceiver might try to run without Tokenizer (Fast doesn't exist) and
with FeatureExtractor some text only pipelines.

* Oops.

* Adding a comment for `update_config_with_model_class`.

* Remove `model_architecture` to get `tiny_config`.

* Finalize rebase.

* Smarter way to handle undefined FastTokenizer.

* Remove old code.

* Addressing some nits.

* Don't instantiate `None`.
This commit is contained in:
Nicolas Patry
2021-12-14 09:43:07 +01:00
committed by GitHub
parent 322d416916
commit 546a91abe9
4 changed files with 83 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
tests/test_pipelines_common.py
View File

@@ -77,12 +77,15 @@ def get_tiny_config_from_class(configuration_class):
model_tester = model_tester_class(parent=None)
if hasattr(model_tester, "get_pipeline_config"):
return model_tester.get_pipeline_config()
config = model_tester.get_pipeline_config()
elif hasattr(model_tester, "get_config"):
return model_tester.get_config()
config = model_tester.get_config()
else:
config = None
logger.warning(f"Model tester {model_tester_class.__name__} has no `get_config()`.")
return config
@lru_cache(maxsize=100)
def get_tiny_tokenizer_from_checkpoint(checkpoint):
@@ -100,11 +103,17 @@ def get_tiny_tokenizer_from_checkpoint(checkpoint):
return tokenizer
def get_tiny_feature_extractor_from_checkpoint(checkpoint, tiny_config):
def get_tiny_feature_extractor_from_checkpoint(checkpoint, tiny_config, feature_extractor_class):
try:
feature_extractor = AutoFeatureExtractor.from_pretrained(checkpoint)
except Exception:
feature_extractor = None
try:
if feature_extractor_class is not None:
feature_extractor = feature_extractor_class()
else:
feature_extractor = None
except Exception:
feature_extractor = None
if hasattr(tiny_config, "image_size") and feature_extractor:
feature_extractor = feature_extractor.__class__(size=tiny_config.image_size, crop_size=tiny_config.image_size)
@@ -168,7 +177,9 @@ class PipelineTestCaseMeta(type):
self.skipTest(f"Ignoring {ModelClass}, cannot create a simple tokenizer")
else:
tokenizer = None
feature_extractor = get_tiny_feature_extractor_from_checkpoint(checkpoint, tiny_config)
feature_extractor = get_tiny_feature_extractor_from_checkpoint(
checkpoint, tiny_config, feature_extractor_class
)
if tokenizer is None and feature_extractor is None:
self.skipTest(
@@ -218,6 +229,13 @@ class PipelineTestCaseMeta(type):
if not tokenizer_classes:
# We need to test even if there are no tokenizers.
tokenizer_classes = [None]
else:
# Remove the non defined tokenizers
# ByT5 and Perceiver are bytes-level and don't define
# FastTokenizer, we can just ignore those.
tokenizer_classes = [
tokenizer_class for tokenizer_class in tokenizer_classes if tokenizer_class is not None
]
for tokenizer_class in tokenizer_classes:
if tokenizer_class is not None: