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

add initial design for uniform processors + align model (#31197)

Browse Source 
* add initial design for uniform processors + align model

* fix mutable default 👀

* add configuration test

* handle structured kwargs w defaults + add test

* protect torch-specific test

* fix style

* fix

* fix assertEqual

* move kwargs merging to processing common

* rework kwargs for type hinting

* just get Unpack from extensions

* run-slow[align]

* handle kwargs passed as nested dict

* add from_pretrained test for nested kwargs handling

* [run-slow]align

* update documentation + imports

* update audio inputs

* protect audio types, silly

* try removing imports

* make things simpler

* simplerer

* move out kwargs test to common mixin

* [run-slow]align

* skip tests for old processors

* [run-slow]align, clip

* !$#@!! protect imports, darn it

* [run-slow]align, clip

* [run-slow]align, clip

* update doc

* improve documentation for default values

* add model_max_length testing

This parameter depends on tokenizers received.

* Raise if kwargs are specified in two places

* fix

* expand VideoInput

* fix

* fix style

* remove defaults values

* add comment to indicate documentation on adding kwargs

* protect imports

* [run-slow]align

* fix

* remove set() that breaks ordering

* test more

* removed unused func

* [run-slow]align
This commit is contained in:
Pablo Montalvo
2024-06-13 16:27:16 +02:00
committed by GitHub
parent 15b3923d65
commit c624d5ba0b
6 changed files with 676 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/models/align/test_processor_align.py
View File

@@ -26,6 +26,8 @@ from transformers.models.bert.tokenization_bert import VOCAB_FILES_NAMES
from transformers.testing_utils import require_vision
from transformers.utils import IMAGE_PROCESSOR_NAME, is_vision_available
from ...test_processing_common import ProcessorTesterMixin
if is_vision_available():
from PIL import Image
@@ -34,7 +36,9 @@ if is_vision_available():
@require_vision
class AlignProcessorTest(unittest.TestCase):
class AlignProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor_class = AlignProcessor
def setUp(self):
self.tmpdirname = tempfile.mkdtemp()
@@ -159,7 +163,6 @@ class AlignProcessorTest(unittest.TestCase):
encoded_processor = processor(text=input_str)
encoded_tok = tokenizer(input_str, padding="max_length", max_length=64)
for key in encoded_tok.keys():
self.assertListEqual(encoded_tok[key], encoded_processor[key])

230
tests/test_processing_common.py
View File

@@ -14,10 +14,19 @@
# limitations under the License.
import inspect
import json
import tempfile
try:
from typing import Unpack
except ImportError:
from typing_extensions import Unpack
import unittest
import numpy as np
from transformers import CLIPTokenizerFast, ProcessorMixin
from transformers.models.auto.processing_auto import processor_class_from_name
from transformers.testing_utils import (
@@ -30,9 +39,13 @@ from transformers.utils import is_vision_available
if is_vision_available():
from PIL import Image
from transformers import CLIPImageProcessor
@require_torch
@require_vision
@require_torch
class ProcessorTesterMixin:
processor_class = None
@@ -64,6 +77,15 @@ class ProcessorTesterMixin:
processor = self.processor_class(**components, **self.prepare_processor_dict())
return processor
@require_vision
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_processor_to_json_string(self):
processor = self.get_processor()
obj = json.loads(processor.to_json_string())
@@ -82,6 +104,214 @@ class ProcessorTesterMixin:
self.assertEqual(processor_second.to_dict(), processor_first.to_dict())
# These kwargs-related tests ensure that processors are correctly instantiated.
# they need to be applied only if an image_processor exists.
def skip_processor_without_typed_kwargs(self, processor):
# TODO this signature check is to test only uniformized processors.
# Once all are updated, remove it.
is_kwargs_typed_dict = False
call_signature = inspect.signature(processor.__call__)
for param in call_signature.parameters.values():
if param.kind == param.VAR_KEYWORD and param.annotation != param.empty:
is_kwargs_typed_dict = (
hasattr(param.annotation, "__origin__") and param.annotation.__origin__ == Unpack
)
if not is_kwargs_typed_dict:
self.skipTest(f"{self.processor_class} doesn't have typed kwargs.")
@require_vision
@require_torch
def test_tokenizer_defaults_preserved_by_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer", max_length=117)
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input, return_tensors="pt")
self.assertEqual(len(inputs["input_ids"][0]), 117)
@require_torch
@require_vision
def test_image_processor_defaults_preserved_by_image_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor", crop_size=(234, 234))
tokenizer = self.get_component("tokenizer", max_length=117)
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertEqual(len(inputs["pixel_values"][0][0]), 234)
@require_vision
@require_torch
def test_kwargs_overrides_default_tokenizer_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer", max_length=117)
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input, return_tensors="pt", max_length=112)
self.assertEqual(len(inputs["input_ids"][0]), 112)
@require_torch
@require_vision
def test_kwargs_overrides_default_image_processor_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor", crop_size=(234, 234))
tokenizer = self.get_component("tokenizer", max_length=117)
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input, crop_size=[224, 224])
self.assertEqual(len(inputs["pixel_values"][0][0]), 224)
@require_torch
@require_vision
def test_unstructured_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer")
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
inputs = processor(
text=input_str,
images=image_input,
return_tensors="pt",
crop_size={"height": 214, "width": 214},
padding="max_length",
max_length=76,
)
self.assertEqual(inputs["pixel_values"].shape[2], 214)
self.assertEqual(len(inputs["input_ids"][0]), 76)
@require_torch
@require_vision
def test_unstructured_kwargs_batched(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer")
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = ["lower newer", "upper older longer string"]
image_input = self.prepare_image_inputs() * 2
inputs = processor(
text=input_str,
images=image_input,
return_tensors="pt",
crop_size={"height": 214, "width": 214},
padding="longest",
max_length=76,
)
self.assertEqual(inputs["pixel_values"].shape[2], 214)
self.assertEqual(len(inputs["input_ids"][0]), 6)
@require_torch
@require_vision
def test_doubly_passed_kwargs(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer")
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = ["lower newer"]
image_input = self.prepare_image_inputs()
with self.assertRaises(ValueError):
_ = processor(
text=input_str,
images=image_input,
images_kwargs={"crop_size": {"height": 222, "width": 222}},
crop_size={"height": 214, "width": 214},
)
@require_torch
@require_vision
def test_structured_kwargs_nested(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer")
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
# Define the kwargs for each modality
all_kwargs = {
"common_kwargs": {"return_tensors": "pt"},
"images_kwargs": {"crop_size": {"height": 214, "width": 214}},
"text_kwargs": {"padding": "max_length", "max_length": 76},
}
inputs = processor(text=input_str, images=image_input, **all_kwargs)
self.skip_processor_without_typed_kwargs(processor)
self.assertEqual(inputs["pixel_values"].shape[2], 214)
self.assertEqual(len(inputs["input_ids"][0]), 76)
@require_torch
@require_vision
def test_structured_kwargs_nested_from_dict(self):
if "image_processor" not in self.processor_class.attributes:
self.skipTest(f"image_processor attribute not present in {self.processor_class}")
image_processor = self.get_component("image_processor")
tokenizer = self.get_component("tokenizer")
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer"
image_input = self.prepare_image_inputs()
# Define the kwargs for each modality
all_kwargs = {
"common_kwargs": {"return_tensors": "pt"},
"images_kwargs": {"crop_size": {"height": 214, "width": 214}},
"text_kwargs": {"padding": "max_length", "max_length": 76},
}
inputs = processor(text=input_str, images=image_input, **all_kwargs)
self.assertEqual(inputs["pixel_values"].shape[2], 214)
self.assertEqual(len(inputs["input_ids"][0]), 76)
class MyProcessor(ProcessorMixin):
attributes = ["image_processor", "tokenizer"]