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

Add support for custom inputs and batched inputs in ProcessorTesterMixin (#33711)

Browse Source 
* add support for custom inputs and batched inputs in ProcessorTesterMixin

* Fix batch_size behavior ProcessorTesterMixin

* Change format prepare inputs batched

* Remove override test pixtral processor

* Remove unnecessary tests and cleanup after new prepare_inputs functions

* Fix instructBlipVideo image processor
This commit is contained in:
Yoni Gozlan
2024-10-01 23:52:03 +02:00
committed by GitHub
parent 1baa08897d
commit 61ac161a9d
8 changed files with 95 additions and 269 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
tests/models/idefics3/test_processing_idefics3.py
View File

@@ -17,6 +17,7 @@ import shutil
import tempfile
import unittest
from io import BytesIO
from typing import Optional
import numpy as np
import requests
@@ -284,44 +285,29 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
)
self.assertEqual(rendered, expected_rendered)
@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")
tokenizer = self.get_component("tokenizer", max_length=117)
# Override as Idefics3Processor needs image tokens in prompts
def prepare_text_inputs(self, batch_size: Optional[int] = None):
if batch_size is None:
return "lower newer <image>"
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
if batch_size < 1:
raise ValueError("batch_size must be greater than 0")
input_str = "lower newer <image>"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertEqual(len(inputs["pixel_values"][0][0]), 3)
self.assertEqual(len(inputs["pixel_values"][0][0][0]), 364) # crop size doesn't affect our image processor
@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", max_image_size={"longest_edge": 32}, size={"longest_edge": 32}
if batch_size == 1:
return ["lower newer <image>"]
return ["lower newer <image>", "<image> upper older longer string"] + ["<image> lower newer"] * (
batch_size - 2
)
tokenizer = self.get_component("tokenizer", max_length=117, padding="max_length")
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor, image_seq_len=2)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer <image>"
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input)
self.assertEqual(len(inputs["pixel_values"][0][0]), 3)
self.assertEqual(len(inputs["pixel_values"][0][0][0]), 32)
self.assertEqual(len(inputs["input_ids"][0]), 117)
# Override as Idefics3Processor needs nested images to work properly with batched inputs
@require_vision
def prepare_image_inputs(self, batch_size: Optional[int] = None):
"""This function prepares a list of PIL images for testing"""
if batch_size is None:
return super().prepare_image_inputs()
if batch_size < 1:
raise ValueError("batch_size must be greater than 0")
return [[super().prepare_image_inputs()]] * batch_size
@require_vision
@require_torch
@@ -333,7 +319,7 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer<image>"
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)
@@ -350,7 +336,7 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer<image>"
input_str = self.prepare_text_inputs()
image_input = self.prepare_image_inputs()
# Define the kwargs for each modality
@@ -378,7 +364,7 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer<image>"
input_str = self.prepare_text_inputs()
image_input = self.prepare_image_inputs()
# Define the kwargs for each modality
@@ -402,7 +388,7 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer<image>"
input_str = self.prepare_text_inputs()
image_input = self.prepare_image_inputs()
inputs = processor(text=input_str, images=image_input, return_tensors="pt")
@@ -419,11 +405,11 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = ["<image>lower newer", "<image>upper older longer string"]
image_input = self.prepare_image_inputs()
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, image_input],
images=image_input,
return_tensors="pt",
padding="longest",
max_length=76,
@@ -446,7 +432,7 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor = self.processor_class(tokenizer=tokenizer, image_processor=image_processor)
self.skip_processor_without_typed_kwargs(processor)
input_str = "lower newer<image>"
input_str = self.prepare_text_inputs()
image_input = self.prepare_image_inputs()
inputs = processor(
text=input_str,