review amy

huggingface · Sep 20, 2024 · 1bbf7ba · 1bbf7ba
1 parent c14da26
commit 1bbf7ba
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 2 deletions.
diff --git a/src/transformers/models/idefics3/modeling_idefics3.py b/src/transformers/models/idefics3/modeling_idefics3.py
@@ -264,7 +264,7 @@ def forward(
         return attn_output, attn_weights
 
 
-# Copied from transformers.models.idefics2.modeling_idefics2.Idefics2VisionFlashAttention2
+# Copied from transformers.models.idefics2.modeling_idefics2.Idefics2VisionFlashAttention2 with Idefics2->Idefics3
 class Idefics3VisionFlashAttention2(Idefics3VisionAttention):
     """
     Idefics3Vision flash attention module. This module inherits from `Idefics3VisionAttention` as the weights of the module stays

diff --git a/tests/models/idefics3/test_image_processing_idefics3.py b/tests/models/idefics3/test_image_processing_idefics3.py
@@ -103,7 +103,9 @@ def get_expected_values(self, image_inputs, batched=False):
 
     def expected_output_image_shape(self, images):
         height, width = self.get_expected_values(images, batched=True)
-        effective_nb_images = self.num_images * 5 if self.do_image_splitting else 1
+        effective_nb_images = (
+            self.num_images * 5 if self.do_image_splitting else 1
+        )  # 5 is a squared image divided into 4 + global image resized
         return effective_nb_images, self.num_channels, height, width
 
     def prepare_image_inputs(