added type hints (#19015)

src/transformers/models/fsmt/modeling_fsmt.py

@@ -468,12 +468,12 @@ class FSMTEncoder(nn.Module):
 
     def forward(
         self,
-        input_ids,
-        attention_mask=None,
-        head_mask=None,
-        output_attentions=False,
-        output_hidden_states=False,
-        return_dict=True,
+        input_ids: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        head_mask: Optional[torch.Tensor] = None,
+        output_attentions: bool = False,
+        output_hidden_states: bool = False,
+        return_dict: bool = True,
     ):
         """
         Args:
@@ -669,18 +669,18 @@ class FSMTDecoder(nn.Module):
 
     def forward(
         self,
-        input_ids,
-        encoder_hidden_states,
-        encoder_padding_mask,
-        decoder_padding_mask,
-        decoder_causal_mask,
-        head_mask=None,
-        cross_attn_head_mask=None,
-        past_key_values=None,
-        use_cache=False,
-        output_attentions=False,
-        output_hidden_states=False,
-        return_dict=True,
+        input_ids: torch.Tensor,
+        encoder_hidden_states: torch.Tensor,
+        encoder_padding_mask: torch.Tensor,
+        decoder_padding_mask: torch.Tensor,
+        decoder_causal_mask: torch.Tensor,
+        head_mask: Optional[torch.Tensor] = None,
+        cross_attn_head_mask: Optional[torch.Tensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        use_cache: bool = False,
+        output_attentions: bool = False,
+        output_hidden_states: bool = False,
+        return_dict: bool = True,
     ):
         """
         Includes several features from "Jointly Learning to Align and Translate with Transformer Models" (Garg et al.,