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

VLMs: patch_size -> num_image_tokens in processing (#33424)

Browse Source 
* use num additional tokens

* fix copies + docs

* another fix copies :)

* add docs

* move order for BC
This commit is contained in:
Raushan Turganbay
2024-11-18 13:21:07 +01:00
committed by GitHub
parent 3ee24e2208
commit 1646ffb4d1
17 changed files with 131 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/models/llava/test_modeling_llava.py
View File

@@ -607,6 +607,7 @@ class LlavaForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.num_additional_image_tokens = 1
processor.patch_size = 14
inputs_expanded = processor(images=raw_image, text=prompt, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 593)
@@ -614,6 +615,7 @@ class LlavaForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(images=raw_image, text=prompt, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs.input_ids.shape[-1] == 18)

4
tests/models/llava_next/test_modeling_llava_next.py
View File

@@ -622,6 +622,7 @@ class LlavaNextForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(text=prompt, images=[raw_image, deer_image], return_tensors="pt").to(
torch_device, torch.float16
)
@@ -630,6 +631,7 @@ class LlavaNextForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(text=prompt, images=[raw_image, deer_image], return_tensors="pt").to(
torch_device, torch.float16
)
@@ -656,12 +658,14 @@ class LlavaNextForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(images=raw_image, text=prompt, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 2356)
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(images=raw_image, text=prompt, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs.input_ids.shape[-1] == 17)

6
tests/models/llava_next_video/test_modeling_llava_next_video.py
View File

@@ -558,12 +558,14 @@ class LlavaNextVideoForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(self.prompt_video, videos=[self.video], return_tensors="pt").to(torch_device)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 1170)
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(self.prompt_video, videos=[self.video], return_tensors="pt").to(torch_device)
self.assertTrue(inputs.input_ids.shape[-1] == 19)
@@ -586,12 +588,14 @@ class LlavaNextVideoForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(self.prompt_image, images=[self.image], return_tensors="pt").to(torch_device)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 2652)
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(self.prompt_image, images=[self.image], return_tensors="pt").to(torch_device)
self.assertTrue(inputs.input_ids.shape[-1] == 19)
@@ -624,6 +628,7 @@ class LlavaNextVideoForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(text=prompt, images=[raw_image, deer_image], return_tensors="pt").to(
torch_device, torch.float16
)
@@ -632,6 +637,7 @@ class LlavaNextVideoForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(text=prompt, images=[raw_image, deer_image], return_tensors="pt").to(
torch_device, torch.float16
)

4
tests/models/video_llava/test_modeling_video_llava.py
View File

@@ -625,12 +625,14 @@ class VideoLlavaForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(prompt, images=image, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 274)
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(prompt, images=image, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs.input_ids.shape[-1] == 19)
@@ -657,12 +659,14 @@ class VideoLlavaForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(prompt, videos=video_file, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 2074)
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(prompt, videos=video_file, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs.input_ids.shape[-1] == 19)

2
tests/models/vipllava/test_modeling_vipllava.py
View File

@@ -374,12 +374,14 @@ class VipLlavaForConditionalGenerationIntegrationTest(unittest.TestCase):
# check processing with expansion of inputs
processor.vision_feature_select_strategy = "default"
processor.patch_size = 14
processor.num_additional_image_tokens = 1
inputs_expanded = processor(prompt, raw_image, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs_expanded.input_ids.shape[-1] == 593)
# check processing without expansion of inputs (legacy behavior)
processor.vision_feature_select_strategy = None
processor.patch_size = None
processor.num_additional_image_tokens = None
inputs = processor(prompt, raw_image, return_tensors="pt").to(torch_device, torch.float16)
self.assertTrue(inputs.input_ids.shape[-1] == 18)