[processor] clean up mulitmodal tests (#37362)
* clkea up mulitmodal processor tests * fixup * fix tests * fix one last test * forgot
This commit is contained in:
Raushan Turganbay
committed by
GitHub
GitHub
parent
3c39c07939
commit
a563999a02
@@ -16,7 +16,6 @@ import shutil
|
||||
import tempfile
|
||||
import unittest
|
||||
from io import BytesIO
|
||||
from typing import Optional
|
||||
|
||||
import numpy as np
|
||||
import requests
|
||||
@@ -41,7 +40,7 @@ class AriaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
|
||||
@classmethod
|
||||
def setUpClass(cls):
|
||||
cls.tmpdirname = tempfile.mkdtemp()
|
||||
processor = AriaProcessor.from_pretrained("m-ric/Aria_hf_2", image_seq_len=2)
|
||||
processor = AriaProcessor.from_pretrained("m-ric/Aria_hf_2", size_conversion={490: 2, 980: 2})
|
||||
processor.save_pretrained(cls.tmpdirname)
|
||||
cls.image1 = Image.open(
|
||||
BytesIO(
|
||||
@@ -74,7 +73,14 @@ class AriaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
|
||||
cls.fake_image_token_id = processor.tokenizer.convert_tokens_to_ids(cls.fake_image_token)
|
||||
cls.global_img_tokens_id = processor.tokenizer(cls.global_img_token, add_special_tokens=False)["input_ids"]
|
||||
cls.padding_token_id = processor.tokenizer.pad_token_id
|
||||
cls.image_seq_len = 256
|
||||
cls.image_seq_len = 2
|
||||
|
||||
@staticmethod
|
||||
def prepare_processor_dict():
|
||||
return {
|
||||
"chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}<|im_start|>{{ message['role'] }}\n{% if message['content'] is string %}{{ message['content'] }}{% elif message['content'] is iterable %}{% for item in message['content'] %}{% if item['type'] == 'text' %}{{ item['text'] }}{% elif item['type'] == 'image' %}<fim_prefix><|img|><fim_suffix>{% endif %}{% endfor %}{% endif %}<|im_end|>\n{% endfor %}{% if add_generation_prompt %}<|im_start|>assistant\n{% endif %}",
|
||||
"size_conversion": {490: 2, 980: 2},
|
||||
} # fmt: skip
|
||||
|
||||
def get_tokenizer(self, **kwargs):
|
||||
return AutoProcessor.from_pretrained(self.tmpdirname, **kwargs).tokenizer
|
||||
@@ -89,24 +95,6 @@ class AriaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
|
||||
def tearDownClass(cls):
|
||||
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
|
||||
|
||||
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}")
|
||||
processor_components = self.prepare_components()
|
||||
processor_components["image_processor"] = self.get_component(
|
||||
"image_processor", do_rescale=True, rescale_factor=1
|
||||
)
|
||||
processor_components["tokenizer"] = self.get_component("tokenizer", max_length=117, padding="max_length")
|
||||
|
||||
processor = self.processor_class(**processor_components)
|
||||
self.skip_processor_without_typed_kwargs(processor)
|
||||
|
||||
input_str = self.prepare_text_inputs()
|
||||
image_input = self.prepare_image_inputs()
|
||||
|
||||
inputs = processor(text=input_str, images=image_input, return_tensors="pt")
|
||||
self.assertLessEqual(inputs[self.images_input_name][0][0].mean(), 0)
|
||||
|
||||
def test_process_interleaved_images_prompts_image_splitting(self):
|
||||
processor = self.get_processor()
|
||||
processor.image_processor.split_image = True
|
||||
@@ -236,155 +224,50 @@ And who is that?<|im_end|>
|
||||
"""
|
||||
self.assertEqual(rendered, expected_rendered)
|
||||
|
||||
# Override as AriaProcessor needs image tokens in prompts
|
||||
def prepare_text_inputs(self, batch_size: Optional[int] = None):
|
||||
if batch_size is None:
|
||||
return "lower newer <|img|>"
|
||||
def test_image_chat_template_accepts_processing_kwargs(self):
|
||||
processor = self.get_processor()
|
||||
if processor.chat_template is None:
|
||||
self.skipTest("Processor has no chat template")
|
||||
|
||||
if batch_size < 1:
|
||||
raise ValueError("batch_size must be greater than 0")
|
||||
messages = [
|
||||
[
|
||||
{
|
||||
"role": "user",
|
||||
"content": [
|
||||
{"type": "text", "text": "What is shown in this image?"},
|
||||
],
|
||||
},
|
||||
]
|
||||
]
|
||||
|
||||
if batch_size == 1:
|
||||
return ["lower newer <|img|>"]
|
||||
return ["lower newer <|img|>", "<|img|> upper older longer string"] + ["<|img|> lower newer"] * (
|
||||
batch_size - 2
|
||||
)
|
||||
|
||||
# Override tests as inputs_ids padded dimension is the second one but not the last one
|
||||
@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=30)
|
||||
|
||||
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
|
||||
self.skip_processor_without_typed_kwargs(processor)
|
||||
input_str = self.prepare_text_inputs()
|
||||
image_input = self.prepare_image_inputs()
|
||||
|
||||
inputs = processor(text=input_str, images=image_input, return_tensors="pt", max_length=30)
|
||||
self.assertEqual(len(inputs["input_ids"][0]), 30)
|
||||
|
||||
@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 = self.prepare_text_inputs()
|
||||
image_input = self.prepare_image_inputs()
|
||||
|
||||
# Define the kwargs for each modality
|
||||
inputs = processor(
|
||||
text=input_str,
|
||||
images=image_input,
|
||||
common_kwargs={"return_tensors": "pt"},
|
||||
images_kwargs={"max_image_size": 980},
|
||||
text_kwargs={"padding": "max_length", "max_length": 120, "truncation": "longest_first"},
|
||||
)
|
||||
self.skip_processor_without_typed_kwargs(processor)
|
||||
|
||||
self.assertEqual(inputs["pixel_values"].shape[3], 980)
|
||||
|
||||
self.assertEqual(len(inputs["input_ids"][0]), 120)
|
||||
|
||||
@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 = self.prepare_text_inputs()
|
||||
image_input = self.prepare_image_inputs()
|
||||
|
||||
# Define the kwargs for each modality
|
||||
all_kwargs = {
|
||||
"common_kwargs": {"return_tensors": "pt"},
|
||||
"images_kwargs": {"max_image_size": 980},
|
||||
"text_kwargs": {"padding": "max_length", "max_length": 120, "truncation": "longest_first"},
|
||||
}
|
||||
|
||||
inputs = processor(text=input_str, images=image_input, **all_kwargs)
|
||||
self.assertEqual(inputs["pixel_values"].shape[3], 980)
|
||||
self.assertEqual(len(inputs["input_ids"][0]), 120)
|
||||
|
||||
@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=30)
|
||||
|
||||
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
|
||||
self.skip_processor_without_typed_kwargs(processor)
|
||||
input_str = self.prepare_text_inputs()
|
||||
image_input = self.prepare_image_inputs()
|
||||
|
||||
inputs = processor(text=input_str, images=image_input, return_tensors="pt")
|
||||
self.assertEqual(len(inputs["input_ids"][0]), 30)
|
||||
|
||||
@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 = self.prepare_text_inputs(batch_size=2)
|
||||
image_input = self.prepare_image_inputs(batch_size=2)
|
||||
inputs = processor(
|
||||
text=input_str,
|
||||
images=image_input,
|
||||
return_tensors="pt",
|
||||
padding="longest",
|
||||
max_length=76,
|
||||
truncation=True,
|
||||
max_image_size=980,
|
||||
)
|
||||
|
||||
self.assertEqual(inputs["pixel_values"].shape[1], 3)
|
||||
self.assertEqual(inputs["pixel_values"].shape[3], 980)
|
||||
self.assertEqual(len(inputs["input_ids"][0]), 76)
|
||||
|
||||
@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 = self.prepare_text_inputs()
|
||||
image_input = self.prepare_image_inputs()
|
||||
inputs = processor(
|
||||
text=input_str,
|
||||
images=image_input,
|
||||
return_tensors="pt",
|
||||
max_image_size=980,
|
||||
formatted_prompt_tokenized = processor.apply_chat_template(
|
||||
messages,
|
||||
add_generation_prompt=True,
|
||||
tokenize=True,
|
||||
padding="max_length",
|
||||
max_length=120,
|
||||
truncation="longest_first",
|
||||
max_length=50,
|
||||
)
|
||||
self.assertEqual(len(formatted_prompt_tokenized[0]), 50)
|
||||
|
||||
self.assertEqual(inputs["pixel_values"].shape[3], 980)
|
||||
self.assertEqual(len(inputs["input_ids"][0]), 120)
|
||||
formatted_prompt_tokenized = processor.apply_chat_template(
|
||||
messages,
|
||||
add_generation_prompt=True,
|
||||
tokenize=True,
|
||||
truncation=True,
|
||||
max_length=5,
|
||||
)
|
||||
self.assertEqual(len(formatted_prompt_tokenized[0]), 5)
|
||||
|
||||
# Now test the ability to return dict
|
||||
messages[0][0]["content"].append(
|
||||
{"type": "image", "url": "https://www.ilankelman.org/stopsigns/australia.jpg"}
|
||||
)
|
||||
out_dict = processor.apply_chat_template(
|
||||
messages,
|
||||
add_generation_prompt=True,
|
||||
tokenize=True,
|
||||
return_dict=True,
|
||||
max_image_size=980,
|
||||
return_tensors="np",
|
||||
)
|
||||
self.assertListEqual(list(out_dict[self.images_input_name].shape), [1, 3, 980, 980])
|
||||
|
||||
Reference in New Issue
Block a user