Fix gradient checkpointing bug in mvp (#21920)

src/transformers/models/mvp/modeling_mvp.py

@@ -1185,6 +1185,13 @@ class MvpDecoder(MvpPreTrainedModel):
             self_attn_prompt = self.self_attn_prompt(prompt_ids)
             cross_attn_prompt = self.cross_attn_prompt(prompt_ids)
 
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
         # decoder layers
         all_hidden_states = () if output_hidden_states else None
         all_self_attns = () if output_attentions else None
@@ -1211,11 +1218,6 @@ class MvpDecoder(MvpPreTrainedModel):
             past_key_value = past_key_values[idx] if past_key_values is not None else None
 
             if self.gradient_checkpointing and self.training:
-                if use_cache:
-                    logger.warning_once(
-                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
-                    )
-                    use_cache = False
 
                 def create_custom_forward(module):
                     def custom_forward(*inputs):