Fix AriaForConditionalGeneration flex attn test (#36604)

AriaForConditionalGeneration depends on idefics3 vision transformer which does not support flex attn

src/transformers/models/aria/modeling_aria.py

@@ -1380,6 +1380,7 @@ ARIA_START_DOCSTRING = r"""
 class AriaForConditionalGeneration(AriaPreTrainedModel, GenerationMixin):
     config_class = AriaConfig
     _supports_flash_attn_2 = False
+    _supports_flex_attn = False
     _supports_sdpa = False
     _tied_weights_keys = ["language_model.lm_head.weight"]
 

src/transformers/models/aria/modular_aria.py

@@ -1348,6 +1348,7 @@ ARIA_START_DOCSTRING = r"""
 class AriaForConditionalGeneration(AriaPreTrainedModel, GenerationMixin):
     config_class = AriaConfig
     _supports_flash_attn_2 = False
+    _supports_flex_attn = False
     _supports_sdpa = False
     _tied_weights_keys = ["language_model.lm_head.weight"]