Qwen: incorrect setup flag (#30776)

qwen does not support the new cache classes

Author: gante

src/transformers/models/qwen2/modeling_qwen2.py

@@ -821,7 +821,6 @@ class Qwen2PreTrainedModel(PreTrainedModel):
     _skip_keys_device_placement = "past_key_values"
     _supports_flash_attn_2 = True
     _supports_sdpa = True
-    _supports_cache_class = True
 
     def _init_weights(self, module):
         std = self.config.initializer_range