Flava multimodal add attention mask (#29446)

* flava multimodal add attn mask

* make style

* check mask is not None

src/transformers/models/flava/modeling_flava.py

@@ -1415,8 +1415,18 @@ class FlavaModel(FlavaPreTrainedModel):
         multimodal_embeddings = None
         multimodal_output = None
         if image_mm_projection is not None and text_mm_projection is not None and not skip_multimodal_encoder:
+            if attention_mask is not None:
+                batch_size, seq_len, _ = image_mm_projection.shape
+                if self.multimodal_model.use_cls_token:
+                    seq_len += 1
+                attention_mask_image = torch.ones(batch_size, seq_len, device=image_mm_projection.device)
+                attention_multimodal = torch.cat([attention_mask_image, attention_mask], dim=1)
+            else:
+                attention_multimodal = None
             multimodal_input = torch.cat([image_mm_projection, text_mm_projection], dim=1)
-            multimodal_output = self.multimodal_model(multimodal_input, return_dict=return_dict)
+            multimodal_output = self.multimodal_model(
+                multimodal_input, attention_mask=attention_multimodal, return_dict=return_dict
+            )
             multimodal_embeddings = multimodal_output[0]
 
         if not return_dict:

tests/models/flava/test_modeling_flava.py

@@ -1287,9 +1287,9 @@ class FlavaModelIntegrationTest(unittest.TestCase):
             outputs = model(**inputs, return_dict=True)
 
         # verify the embeddings
-        self.assertAlmostEqual(outputs.image_embeddings.sum().item(), -1352.53540, places=4)
+        self.assertAlmostEqual(outputs.image_embeddings.sum().item(), -1352.54943, places=4)
         self.assertAlmostEqual(outputs.text_embeddings.sum().item(), -198.98225, places=4)
-        self.assertAlmostEqual(outputs.multimodal_embeddings.sum().item(), -3988.51367, places=4)
+        self.assertAlmostEqual(outputs.multimodal_embeddings.sum().item(), -4030.466552, places=4)
 
 
 @require_vision
@@ -1339,9 +1339,9 @@ class FlavaForPreTrainingIntegrationTest(unittest.TestCase):
 
         expected_logits = torch.tensor([[16.1291, 8.4033], [16.1291, 8.4033]], device=torch_device)
         self.assertTrue(torch.allclose(outputs.contrastive_logits_per_image, expected_logits, atol=1e-3))
-        self.assertAlmostEqual(outputs.loss_info.mmm_text.item(), 1.75533199, places=4)
-        self.assertAlmostEqual(outputs.loss_info.mmm_image.item(), 7.0290069, places=4)
-        self.assertAlmostEqual(outputs.loss.item(), 11.0626, places=4)
+        self.assertAlmostEqual(outputs.loss_info.mmm_text.item(), 2.0736470, places=4)
+        self.assertAlmostEqual(outputs.loss_info.mmm_image.item(), 7.025580, places=4)
+        self.assertAlmostEqual(outputs.loss.item(), 11.37761, places=4)
 
     @slow
     def test_inference_with_itm_labels(self):
@@ -1390,6 +1390,6 @@ class FlavaForPreTrainingIntegrationTest(unittest.TestCase):
 
         expected_logits = torch.tensor([[16.1291, 8.4033], [16.1291, 8.4033]], device=torch_device)
         self.assertTrue(torch.allclose(outputs.contrastive_logits_per_image, expected_logits, atol=1e-3))
-        self.assertAlmostEqual(outputs.loss_info.mmm_text.item(), 1.75533199, places=4)
-        self.assertAlmostEqual(outputs.loss_info.mmm_image.item(), 6.89590501, places=4)
-        self.assertAlmostEqual(outputs.loss.item(), 9.1995, places=4)
+        self.assertAlmostEqual(outputs.loss_info.mmm_text.item(), 2.0736470, places=4)
+        self.assertAlmostEqual(outputs.loss_info.mmm_image.item(), 6.8962264, places=4)
+        self.assertAlmostEqual(outputs.loss.item(), 9.6090, places=4)