Pass missing arguments to `SeamlessM4Tv2ConformerEncoderLayer.forward()` when...

Pass missing arguments to `SeamlessM4Tv2ConformerEncoderLayer.forward()` when gradient checkpointing is enabled (#31945)

* pass missing arguments when gradient checkpointing is enabled for SeamlessM4Tv2

* fix same bug in SeamlessM4Tv1

* pass args, not kwargs

src/transformers/models/seamless_m4t/modeling_seamless_m4t.py

@@ -842,6 +842,8 @@ class SeamlessM4TConformerEncoder(nn.Module):
                         hidden_states,
                         attention_mask,
                         relative_position_embeddings,
+                        output_attentions,
+                        conv_attention_mask,
                     )
                 else:
                     layer_outputs = layer(

src/transformers/models/seamless_m4t_v2/modeling_seamless_m4t_v2.py

@@ -788,6 +788,8 @@ class SeamlessM4Tv2ConformerEncoder(nn.Module):
                         layer.__call__,
                         hidden_states,
                         attention_mask,
+                        output_attentions,
+                        conv_attention_mask,
                     )
                 else:
                     layer_outputs = layer(