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

Allow FP16 or other precision inference for Pipelines (#31342)

Browse Source 
* cast image features to model.dtype where needed to support FP16 or other precision in pipelines

* Update src/transformers/pipelines/image_feature_extraction.py

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Use .to instead

* Add FP16 pipeline support for zeroshot audio classification

* Remove unused torch imports

* Add docs on FP16 pipeline

* Remove unused import

* Add FP16 tests to pipeline mixin

* Add fp16 placeholder for mask_generation pipeline test

* Add FP16 tests for all pipelines

* Fix formatting

* Remove torch_dtype arg from is_pipeline_test_to_skip*

* Fix format

* trigger ci

---------

Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
This commit is contained in:
Billy Cao
2024-07-06 00:21:50 +08:00
committed by GitHub
parent e786844425
commit ac26260436
45 changed files with 354 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/pipelines/test_pipelines_audio_classification.py
View File

@@ -35,8 +35,10 @@ class AudioClassificationPipelineTests(unittest.TestCase):
model_mapping = MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING
tf_model_mapping = TF_MODEL_FOR_AUDIO_CLASSIFICATION_MAPPING
def get_test_pipeline(self, model, tokenizer, processor):
audio_classifier = AudioClassificationPipeline(model=model, feature_extractor=processor)
def get_test_pipeline(self, model, tokenizer, processor, torch_dtype="float32"):
audio_classifier = AudioClassificationPipeline(
model=model, feature_extractor=processor, torch_dtype=torch_dtype
)
# test with a raw waveform
audio = np.zeros((34000,))