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

Add Idefics2/3 and SmolVLM Fast image processors + improvements for fast image processors (#38157)

Browse Source 
* add working idefics2 fast and improvements for fast nested images processing

* add fast image processors idefics 3 and smolvlm

* cleanup tests

* fic doc idefics2

* PR review and fix issues after merge

* Force providing disable_grouping to group_images_by_shape

* simplify group_images_by_shape

* fix modular

* Fix nits after review
This commit is contained in:
Yoni Gozlan
2025-06-23 10:17:25 -04:00
committed by GitHub
parent 1a96127e46
commit d29482cc91
61 changed files with 2023 additions and 425 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
tests/test_image_processing_common.py
View File

@@ -162,6 +162,10 @@ class ImageProcessingTestMixin:
self.image_processor_list = image_processor_list
def _assert_slow_fast_tensors_equivalence(self, slow_tensor, fast_tensor, atol=1e-1, rtol=1e-3, mean_atol=5e-3):
torch.testing.assert_close(slow_tensor, fast_tensor, atol=atol, rtol=rtol)
self.assertLessEqual(torch.mean(torch.abs(slow_tensor - fast_tensor)).item(), mean_atol)
@require_vision
@require_torch
def test_slow_fast_equivalence(self):
@@ -179,10 +183,7 @@ class ImageProcessingTestMixin:
encoding_slow = image_processor_slow(dummy_image, return_tensors="pt")
encoding_fast = image_processor_fast(dummy_image, return_tensors="pt")
torch.testing.assert_close(encoding_slow.pixel_values, encoding_fast.pixel_values, atol=1e-1, rtol=1e-3)
self.assertLessEqual(
torch.mean(torch.abs(encoding_slow.pixel_values - encoding_fast.pixel_values)).item(), 5e-3
)
self._assert_slow_fast_tensors_equivalence(encoding_slow.pixel_values, encoding_fast.pixel_values)
@require_vision
@require_torch
@@ -205,10 +206,7 @@ class ImageProcessingTestMixin:
encoding_slow = image_processor_slow(dummy_images, return_tensors="pt")
encoding_fast = image_processor_fast(dummy_images, return_tensors="pt")
torch.testing.assert_close(encoding_slow.pixel_values, encoding_fast.pixel_values, atol=1e-1, rtol=1e-3)
self.assertLessEqual(
torch.mean(torch.abs(encoding_slow.pixel_values - encoding_fast.pixel_values)).item(), 5e-3
)
self._assert_slow_fast_tensors_equivalence(encoding_slow.pixel_values, encoding_fast.pixel_values)
@require_vision
@require_torch
@@ -577,8 +575,10 @@ class ImageProcessingTestMixin:
image_processor = torch.compile(image_processor, mode="reduce-overhead")
output_compiled = image_processor(input_image, device=torch_device, return_tensors="pt")
torch.testing.assert_close(output_eager.pixel_values, output_compiled.pixel_values, rtol=1e-4, atol=1e-4)
print(output_eager.pixel_values.dtype, output_compiled.pixel_values.dtype)
self._assert_slow_fast_tensors_equivalence(
output_eager.pixel_values, output_compiled.pixel_values, atol=1e-4, rtol=1e-4, mean_atol=1e-5
)
class AnnotationFormatTestMixin: