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

[processor] Add 'model input names' property (#20117)

Browse Source 
* [processor] Add 'model input names' property

* add test

* no f string

* add generic property method to mixin

* copy to multimodal

* copy to vision

* tests for all audio

* remove ad-hoc tests

* style

* fix flava test

* fix test

* fix processor code
This commit is contained in:
Sanchit Gandhi
2022-11-10 19:29:20 +00:00
committed by GitHub
parent 68187c4642
commit 905e5773a3
24 changed files with 261 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
tests/models/clip/test_processor_clip.py
View File

@@ -187,3 +187,16 @@ class CLIPProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = CLIPProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)

13
tests/models/flava/test_processor_flava.py
View File

@@ -231,3 +231,16 @@ class FlavaProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = FlavaProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)

27
tests/models/layoutlmv2/test_processor_layoutlmv2.py
View File

@@ -19,6 +19,8 @@ import tempfile
import unittest
from typing import List
import numpy as np
from transformers import PreTrainedTokenizer, PreTrainedTokenizerBase, PreTrainedTokenizerFast
from transformers.models.layoutlmv2 import LayoutLMv2Tokenizer, LayoutLMv2TokenizerFast
from transformers.models.layoutlmv2.tokenization_layoutlmv2 import VOCAB_FILES_NAMES
@@ -86,6 +88,17 @@ class LayoutLMv2ProcessorTest(unittest.TestCase):
def tearDown(self):
shutil.rmtree(self.tmpdirname)
def prepare_image_inputs(self):
"""This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
or a list of PyTorch tensors if one specifies torchify=True.
"""
image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]
image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
return image_inputs
def test_save_load_pretrained_default(self):
feature_extractor = self.get_feature_extractor()
tokenizers = self.get_tokenizers()
@@ -133,6 +146,20 @@ class LayoutLMv2ProcessorTest(unittest.TestCase):
self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
self.assertIsInstance(processor.feature_extractor, LayoutLMv2FeatureExtractor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = LayoutLMv2Processor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
# add extra args
inputs = processor(text=input_str, images=image_input, return_codebook_pixels=False, return_image_mask=False)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)
@slow
def test_overflowing_tokens(self):
# In the case of overflowing tokens, test that we still have 1-to-1 mapping between the images and input_ids (sequences that are too long are broken down into multiple sequences).

27
tests/models/layoutlmv3/test_processor_layoutlmv3.py
View File

@@ -19,6 +19,8 @@ import tempfile
import unittest
from typing import List
import numpy as np
from transformers import PreTrainedTokenizer, PreTrainedTokenizerBase, PreTrainedTokenizerFast
from transformers.models.layoutlmv3 import LayoutLMv3Tokenizer, LayoutLMv3TokenizerFast
from transformers.models.layoutlmv3.tokenization_layoutlmv3 import VOCAB_FILES_NAMES
@@ -99,6 +101,17 @@ class LayoutLMv3ProcessorTest(unittest.TestCase):
def tearDown(self):
shutil.rmtree(self.tmpdirname)
def prepare_image_inputs(self):
"""This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
or a list of PyTorch tensors if one specifies torchify=True.
"""
image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]
image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
return image_inputs
def test_save_load_pretrained_default(self):
feature_extractor = self.get_feature_extractor()
tokenizers = self.get_tokenizers()
@@ -146,6 +159,20 @@ class LayoutLMv3ProcessorTest(unittest.TestCase):
self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
self.assertIsInstance(processor.feature_extractor, LayoutLMv3FeatureExtractor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = LayoutLMv3Processor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
# add extra args
inputs = processor(text=input_str, images=image_input, return_codebook_pixels=False, return_image_mask=False)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)
# different use cases tests
@require_torch

27
tests/models/layoutxlm/test_processor_layoutxlm.py
View File

@@ -19,6 +19,8 @@ import tempfile
import unittest
from typing import List
import numpy as np
from transformers import PreTrainedTokenizer, PreTrainedTokenizerBase, PreTrainedTokenizerFast
from transformers.models.layoutxlm import LayoutXLMTokenizer, LayoutXLMTokenizerFast
from transformers.testing_utils import (
@@ -74,6 +76,17 @@ class LayoutXLMProcessorTest(unittest.TestCase):
def tearDown(self):
shutil.rmtree(self.tmpdirname)
def prepare_image_inputs(self):
"""This function prepares a list of PIL images, or a list of numpy arrays if one specifies numpify=True,
or a list of PyTorch tensors if one specifies torchify=True.
"""
image_inputs = [np.random.randint(255, size=(3, 30, 400), dtype=np.uint8)]
image_inputs = [Image.fromarray(np.moveaxis(x, 0, -1)) for x in image_inputs]
return image_inputs
def test_save_load_pretrained_default(self):
feature_extractor = self.get_feature_extractor()
tokenizers = self.get_tokenizers()
@@ -126,6 +139,20 @@ class LayoutXLMProcessorTest(unittest.TestCase):
self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
self.assertIsInstance(processor.feature_extractor, LayoutLMv2FeatureExtractor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = LayoutXLMProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
# add extra args
inputs = processor(text=input_str, images=image_input, return_codebook_pixels=False, return_image_mask=False)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)
@slow
def test_overflowing_tokens(self):
# In the case of overflowing tokens, test that we still have 1-to-1 mapping between the images and input_ids (sequences that are too long are broken down into multiple sequences).

12
tests/models/markuplm/test_processor_markuplm.py
View File

@@ -133,6 +133,18 @@ class MarkupLMProcessorTest(unittest.TestCase):
self.assertEqual(processor.feature_extractor.to_json_string(), feature_extractor_add_kwargs.to_json_string())
self.assertIsInstance(processor.feature_extractor, MarkupLMFeatureExtractor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = MarkupLMProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
self.assertListEqual(
processor.model_input_names,
tokenizer.model_input_names,
msg="`processor` and `tokenizer` model input names do not match",
)
# different use cases tests
@require_bs4

12
tests/models/mctct/test_processor_mctct.py
View File

@@ -144,3 +144,15 @@ class MCTCTProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = MCTCTProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
self.assertListEqual(
processor.model_input_names,
feature_extractor.model_input_names,
msg="`processor` and `feature_extractor` model input names do not match",
)

13
tests/models/owlvit/test_processor_owlvit.py
View File

@@ -239,3 +239,16 @@ class OwlViTProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = OwlViTProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)

12
tests/models/speech_to_text/test_processor_speech_to_text.py
View File

@@ -144,3 +144,15 @@ class Speech2TextProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = Speech2TextProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
self.assertListEqual(
processor.model_input_names,
feature_extractor.model_input_names,
msg="`processor` and `feature_extractor` model input names do not match",
)

13
tests/models/vision_text_dual_encoder/test_processor_vision_text_dual_encoder.py
View File

@@ -168,3 +168,16 @@ class VisionTextDualEncoderProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_image_processor()
tokenizer = self.get_tokenizer()
processor = VisionTextDualEncoderProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertListEqual(list(inputs.keys()), processor.model_input_names)

12
tests/models/wav2vec2/test_processor_wav2vec2.py
View File

@@ -137,3 +137,15 @@ class Wav2Vec2ProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = Wav2Vec2Processor(tokenizer=tokenizer, feature_extractor=feature_extractor)
self.assertListEqual(
processor.model_input_names,
feature_extractor.model_input_names,
msg="`processor` and `feature_extractor` model input names do not match",
)

13
tests/models/wav2vec2_with_lm/test_processor_wav2vec2_with_lm.py
View File

@@ -367,6 +367,19 @@ class Wav2Vec2ProcessorWithLMTest(unittest.TestCase):
self.assertListEqual(decoded_wav2vec2.text, decoded_auto.text)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
decoder = self.get_decoder()
processor = Wav2Vec2ProcessorWithLM(tokenizer=tokenizer, feature_extractor=feature_extractor, decoder=decoder)
self.assertListEqual(
processor.model_input_names,
feature_extractor.model_input_names,
msg="`processor` and `feature_extractor` model input names do not match",
)
@staticmethod
def get_from_offsets(offsets, key):
retrieved_list = [d[key] for d in offsets]

12
tests/models/whisper/test_processor_whisper.py
View File

@@ -116,3 +116,15 @@ class WhisperProcessorTest(unittest.TestCase):
decoded_tok = tokenizer.batch_decode(predicted_ids)
self.assertListEqual(decoded_tok, decoded_processor)
def test_model_input_names(self):
feature_extractor = self.get_feature_extractor()
tokenizer = self.get_tokenizer()
processor = WhisperProcessor(tokenizer=tokenizer, feature_extractor=feature_extractor)
self.assertListEqual(
processor.model_input_names,
feature_extractor.model_input_names,
msg="`processor` and `feature_extractor` model input names do not match",
)