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

[processors] add tests for helper fn (#39629)

Browse Source 
* add tests for helpers

* duplicate test for each model

* why llava next video has no helper

* oops must have been in the commit

* fix test after rebase

* add copy from
This commit is contained in:
Raushan Turganbay
2025-07-28 11:41:58 +02:00
committed by GitHub
parent 6638b3642d
commit 8b237b8639
40 changed files with 454 additions and 58 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/models/aria/test_image_processing_aria.py
View File

@@ -302,3 +302,19 @@ class AriaImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
encoded_images.shape[:-1] if input_data_format == ChannelDimension.LAST else encoded_images.shape[1:]
)
self.assertEqual(encoded_image_shape, image_shape)
def test_get_num_patches_without_images(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
num_patches = image_processing.get_number_of_image_patches(height=100, width=100, images_kwargs={})
self.assertEqual(num_patches, 1)
num_patches = image_processing.get_number_of_image_patches(
height=300, width=500, images_kwargs={"split_image": True}
)
self.assertEqual(num_patches, 1)
num_patches = image_processing.get_number_of_image_patches(
height=100, width=100, images_kwargs={"split_image": True, "max_image_size": 200}
)
self.assertEqual(num_patches, 19)

13
tests/models/aria/test_processor_aria.py
View File

@@ -95,6 +95,19 @@ class AriaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def test_process_interleaved_images_prompts_image_splitting(self):
processor = self.get_processor()
processor.image_processor.split_image = True

13
tests/models/aya_vision/test_processor_aya_vision.py
View File

@@ -80,6 +80,19 @@ class AyaVisionProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
@require_torch
def test_process_interleaved_images_videos(self):
processor = self.get_processor()

13
tests/models/chameleon/test_processor_chameleon.py
View File

@@ -74,3 +74,16 @@ class ChameleonProcessorTest(ProcessorTesterMixin, unittest.TestCase):
@staticmethod
def prepare_processor_dict():
return {"image_seq_length": 2} # fmt: skip
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)

13
tests/models/colpali/test_processing_colpali.py
View File

@@ -54,6 +54,19 @@ class ColPaliProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
@require_torch
@require_vision
def test_process_images(self):

13
tests/models/colqwen2/test_processing_colqwen2.py
View File

@@ -57,6 +57,19 @@ class ColQwen2ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def test_process_images(self):
# Processor configuration
image_input = self.prepare_image_inputs()

13
tests/models/emu3/test_processor_emu3.py
View File

@@ -90,3 +90,16 @@ class Emu3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
# For an image where pixels go from 0 to 255 the diff can be 1 due to some numerical precision errors when scaling and unscaling
self.assertTrue(np.abs(orig_image - unnormalized_images).max() >= 1)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)

13
tests/models/fuyu/test_processor_fuyu.py
View File

@@ -64,6 +64,19 @@ class FuyuProcessingTest(ProcessorTesterMixin, unittest.TestCase):
def get_image_processor(self, **kwargs):
return AutoProcessor.from_pretrained(self.tmpdirname, **kwargs).image_processor
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def test_fuyu_processing(self):
"""
Test to ensure that the standard processing on a gold example matches adept's code.

13
tests/models/gemma3/test_processing_gemma3.py
View File

@@ -58,6 +58,19 @@ class Gemma3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
processor.save_pretrained(cls.tmpdirname)
cls.image_token = processor.boi_token
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
@classmethod
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)

21
tests/models/got_ocr2/test_image_processing_got_ocr2.py
View File

@@ -169,3 +169,24 @@ class GotOcr2ProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
)
self.assertEqual(len(processed_images[0]), 5)
self.assertEqual(processed_images.shape[-2:], (20, 20))
def test_get_num_patches_without_images(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
num_patches = image_processing.get_number_of_image_patches(height=100, width=100, images_kwargs={})
self.assertEqual(num_patches, 1)
num_patches = image_processing.get_number_of_image_patches(
height=300, width=500, images_kwargs={"crop_to_patches": False}
)
self.assertEqual(num_patches, 1)
num_patches = image_processing.get_number_of_image_patches(
height=100, width=100, images_kwargs={"crop_to_patches": True}
)
self.assertEqual(num_patches, 10)
num_patches = image_processing.get_number_of_image_patches(
height=100, width=100, images_kwargs={"crop_to_patches": True, "max_patches": 200}
)
self.assertEqual(num_patches, 50)

25
tests/models/idefics3/test_image_processing_idefics3.py
View File

@@ -358,3 +358,28 @@ class Idefics3ImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
)
self.assertEqual(encoding_slow.rows, encoding_fast.rows)
self.assertEqual(encoding_slow.cols, encoding_fast.cols)
def test_get_num_patches_without_images(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=100, width=100, images_kwargs={}
)
self.assertEqual(num_patches_and_row_cols, (5, 2, 2))
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=300, width=500, images_kwargs={"do_image_splitting": False}
)
self.assertEqual(num_patches_and_row_cols, (1, 1, 1))
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=300, width=500, images_kwargs={"do_image_splitting": True}
)
self.assertEqual(num_patches_and_row_cols, (5, 2, 2))
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=300,
width=600,
images_kwargs={"do_image_splitting": True, "max_image_size": {"longest_edge": 30}},
)
self.assertEqual(num_patches_and_row_cols, (3, 1, 2))

13
tests/models/idefics3/test_processor_idefics3.py
View File

@@ -84,6 +84,19 @@ class Idefics3ProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def prepare_processor_dict():
return {"image_seq_len": 2}
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def get_split_image_expected_tokens(self, processor, image_rows, image_cols):
text_split_images = []
for n_h in range(image_rows):

13
tests/models/internvl/test_processor_internvl.py
View File

@@ -97,6 +97,19 @@ class InternVLProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
@require_av
@require_torch
def test_process_interleaved_images_videos(self):

12
tests/models/llava/test_processor_llava.py
View File

@@ -61,6 +61,18 @@ class LlavaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
"vision_feature_select_strategy": "default"
} # fmt: skip
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def test_chat_template_is_saved(self):
processor_loaded = self.processor_class.from_pretrained(self.tmpdirname)
processor_dict_loaded = json.loads(processor_loaded.to_json_string())

13
tests/models/llava_next/test_processor_llava_next.py
View File

@@ -66,6 +66,19 @@ class LlavaNextProcessorTest(ProcessorTesterMixin, unittest.TestCase):
"vision_feature_select_strategy": "default"
} # fmt: skip
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_chat_template_is_saved
def test_chat_template_is_saved(self):
processor_loaded = self.processor_class.from_pretrained(self.tmpdirname)

13
tests/models/llava_next_video/test_processor_llava_next_video.py
View File

@@ -75,6 +75,19 @@ class LlavaNextVideoProcessorTest(ProcessorTesterMixin, unittest.TestCase):
"vision_feature_select_strategy": "default",
}
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_chat_template_is_saved
def test_chat_template_is_saved(self):
processor_loaded = self.processor_class.from_pretrained(self.tmpdirname)

13
tests/models/llava_onevision/test_processor_llava_onevision.py
View File

@@ -79,6 +79,19 @@ class LlavaOnevisionProcessorTest(ProcessorTesterMixin, unittest.TestCase):
"vision_feature_select_strategy": "default"
} # fmt: skip
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_chat_template_is_saved
def test_chat_template_is_saved(self):
processor_loaded = self.processor_class.from_pretrained(self.tmpdirname)

13
tests/models/paligemma/test_processor_paligemma.py
View File

@@ -48,6 +48,19 @@ class PaliGemmaProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
@require_torch
@require_vision
def test_image_seq_length(self):

13
tests/models/qwen2_5_vl/test_processor_qwen2_5_vl.py
View File

@@ -65,6 +65,19 @@ class Qwen2_5_VLProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def test_save_load_pretrained_default(self):
tokenizer = self.get_tokenizer()
image_processor = self.get_image_processor()

14
tests/models/qwen2_vl/test_image_processing_qwen2_vl.py
View File

@@ -394,3 +394,17 @@ class Qwen2VLImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
self._assert_slow_fast_tensors_equivalence(
encoding_slow.image_grid_thw.float(), encoding_fast.image_grid_thw.float()
)
def test_get_num_patches_without_images(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
num_patches = image_processing.get_number_of_image_patches(height=100, width=100, images_kwargs={})
self.assertEqual(num_patches, 64)
num_patches = image_processing.get_number_of_image_patches(height=200, width=50, images_kwargs={})
self.assertEqual(num_patches, 56)
num_patches = image_processing.get_number_of_image_patches(
height=100, width=100, images_kwargs={"patch_size": 28}
)
self.assertEqual(num_patches, 16)

13
tests/models/qwen2_vl/test_processor_qwen2_vl.py
View File

@@ -68,6 +68,19 @@ class Qwen2VLProcessorTest(ProcessorTesterMixin, unittest.TestCase):
def tearDownClass(cls):
shutil.rmtree(cls.tmpdirname, ignore_errors=True)
# Copied from tests.models.llava.test_processor_llava.LlavaProcessorTest.test_get_num_vision_tokens
def test_get_num_vision_tokens(self):
"Tests general functionality of the helper used internally in vLLM"
processor = self.get_processor()
output = processor._get_num_multimodal_tokens(image_sizes=[(100, 100), (300, 100), (500, 30)])
self.assertTrue("num_image_tokens" in output)
self.assertEqual(len(output["num_image_tokens"]), 3)
self.assertTrue("num_image_patches" in output)
self.assertEqual(len(output["num_image_patches"]), 3)
def test_save_load_pretrained_default(self):
tokenizer = self.get_tokenizer()
image_processor = self.get_image_processor()

25
tests/models/smolvlm/test_image_processing_smolvlm.py
View File

@@ -358,3 +358,28 @@ class SmolVLMImageProcessingTest(ImageProcessingTestMixin, unittest.TestCase):
)
self.assertEqual(encoding_slow.rows, encoding_fast.rows)
self.assertEqual(encoding_slow.cols, encoding_fast.cols)
def test_get_num_patches_without_images(self):
for image_processing_class in self.image_processor_list:
image_processing = image_processing_class(**self.image_processor_dict)
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=100, width=100, images_kwargs={}
)
self.assertEqual(num_patches_and_row_cols, (5, 2, 2))
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=300, width=500, images_kwargs={"do_image_splitting": False}
)
self.assertEqual(num_patches_and_row_cols, (1, 1, 1))
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=300, width=500, images_kwargs={"do_image_splitting": True}
)
self.assertEqual(num_patches_and_row_cols, (5, 2, 2))
num_patches_and_row_cols = image_processing.get_number_of_image_patches(
height=300,
width=600,
images_kwargs={"do_image_splitting": True, "max_image_size": {"longest_edge": 30}},
)
self.assertEqual(num_patches_and_row_cols, (3, 1, 2))