Patch YOLOS and others (#29353)

Fix issue

src/transformers/models/conditional_detr/modeling_conditional_detr.py

@@ -2514,9 +2514,10 @@ class ConditionalDetrLoss(nn.Module):
         num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device)
 
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_boxes = reduce(num_boxes)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_boxes = reduce(num_boxes)
+                world_size = PartialState().num_processes
         num_boxes = torch.clamp(num_boxes / world_size, min=1).item()
 
         # Compute all the requested losses

src/transformers/models/deformable_detr/modeling_deformable_detr.py

@@ -2282,9 +2282,10 @@ class DeformableDetrLoss(nn.Module):
         num_boxes = sum(len(t["class_labels"]) for t in targets)
         num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_boxes = reduce(num_boxes)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_boxes = reduce(num_boxes)
+                world_size = PartialState().num_processes
         num_boxes = torch.clamp(num_boxes / world_size, min=1).item()
 
         # Compute all the requested losses

src/transformers/models/deta/modeling_deta.py

@@ -2345,9 +2345,10 @@ class DetaLoss(nn.Module):
         num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device)
         # Check that we have initialized the distributed state
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_boxes = reduce(num_boxes)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_boxes = reduce(num_boxes)
+                world_size = PartialState().num_processes
         num_boxes = torch.clamp(num_boxes / world_size, min=1).item()
 
         # Compute all the requested losses

src/transformers/models/detr/modeling_detr.py

@@ -2210,9 +2210,10 @@ class DetrLoss(nn.Module):
         num_boxes = sum(len(t["class_labels"]) for t in targets)
         num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_boxes = reduce(num_boxes)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_boxes = reduce(num_boxes)
+                world_size = PartialState().num_processes
         num_boxes = torch.clamp(num_boxes / world_size, min=1).item()
 
         # Compute all the requested losses

src/transformers/models/mask2former/modeling_mask2former.py

@@ -791,14 +791,15 @@ class Mask2FormerLoss(nn.Module):
         Computes the average number of target masks across the batch, for normalization purposes.
         """
         num_masks = sum([len(classes) for classes in class_labels])
-        num_masks_pt = torch.as_tensor(num_masks, dtype=torch.float, device=device)
+        num_masks = torch.as_tensor(num_masks, dtype=torch.float, device=device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_masks_pt = reduce(num_masks_pt)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_masks = reduce(num_masks)
+                world_size = PartialState().num_processes
 
-        num_masks_pt = torch.clamp(num_masks_pt / world_size, min=1)
-        return num_masks_pt
+        num_masks = torch.clamp(num_masks / world_size, min=1)
+        return num_masks
 
 
 # Copied from transformers.models.deformable_detr.modeling_deformable_detr.multi_scale_deformable_attention

src/transformers/models/maskformer/modeling_maskformer.py

@@ -1198,14 +1198,15 @@ class MaskFormerLoss(nn.Module):
         Computes the average number of target masks across the batch, for normalization purposes.
         """
         num_masks = sum([len(classes) for classes in class_labels])
-        num_masks_pt = torch.as_tensor(num_masks, dtype=torch.float, device=device)
+        num_masks = torch.as_tensor(num_masks, dtype=torch.float, device=device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_masks_pt = reduce(num_masks_pt)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_masks = reduce(num_masks)
+                world_size = PartialState().num_processes
 
-        num_masks_pt = torch.clamp(num_masks_pt / world_size, min=1)
-        return num_masks_pt
+        num_masks = torch.clamp(num_masks / world_size, min=1)
+        return num_masks
 
 
 class MaskFormerFPNConvLayer(nn.Module):

src/transformers/models/oneformer/modeling_oneformer.py

@@ -727,14 +727,15 @@ class OneFormerLoss(nn.Module):
         Computes the average number of target masks across the batch, for normalization purposes.
         """
         num_masks = sum([len(classes) for classes in class_labels])
-        num_masks_pt = torch.as_tensor([num_masks], dtype=torch.float, device=device)
+        num_masks = torch.as_tensor([num_masks], dtype=torch.float, device=device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_masks_pt = reduce(num_masks_pt)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_masks = reduce(num_masks)
+                world_size = PartialState().num_processes
 
-        num_masks_pt = torch.clamp(num_masks_pt / world_size, min=1)
-        return num_masks_pt
+        num_masks = torch.clamp(num_masks / world_size, min=1)
+        return num_masks
 
 
 @dataclass

src/transformers/models/table_transformer/modeling_table_transformer.py

@@ -1757,9 +1757,10 @@ class TableTransformerLoss(nn.Module):
         num_boxes = sum(len(t["class_labels"]) for t in targets)
         num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_boxes = reduce(num_boxes)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_boxes = reduce(num_boxes)
+                world_size = PartialState().num_processes
         num_boxes = torch.clamp(num_boxes / world_size, min=1).item()
 
         # Compute all the requested losses

src/transformers/models/yolos/modeling_yolos.py

@@ -1079,9 +1079,10 @@ class YolosLoss(nn.Module):
         num_boxes = sum(len(t["class_labels"]) for t in targets)
         num_boxes = torch.as_tensor([num_boxes], dtype=torch.float, device=next(iter(outputs.values())).device)
         world_size = 1
-        if PartialState._shared_state != {}:
-            num_boxes = reduce(num_boxes)
-            world_size = PartialState().num_processes
+        if is_accelerate_available():
+            if PartialState._shared_state != {}:
+                num_boxes = reduce(num_boxes)
+                world_size = PartialState().num_processes
         num_boxes = torch.clamp(num_boxes / world_size, min=1).item()
 
         # Compute all the requested losses