Update modeling_llama.py

reverting to prev version

modeling_llama.py

@@ -985,8 +985,7 @@ class LlamaModel(LlamaPreTrainedModel):
             if past_key_values is not None and not isinstance(
                 past_key_values, StaticCache
             ):
-                if not isinstance(past_key_values, DynamicCache):
-                    past_key_values = DynamicCache.from_legacy_cache(past_key_values)
+                past_key_values = DynamicCache.from_legacy_cache(past_key_values)
                 past_seen_tokens = past_key_values.get_seq_length()
 
         if cache_position is None:
@@ -1057,7 +1056,11 @@ class LlamaModel(LlamaPreTrainedModel):
 
         next_cache = None
         if use_cache:
-            next_cache = next_decoder_cache
+            next_cache = (
+                next_decoder_cache.to_legacy_cache()
+                if isinstance(next_decoder_cache, Cache)
+                else next_decoder_cache
+            )
         if not return_dict:
             return tuple(
                 v