remove fa

modeling_audio.py

@@ -91,7 +91,6 @@ class WhisperRoPEFlashAttn(WhisperFlashAttention2):
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
         # WhisperFlashAttention2 attention does not support output_attentions
         if output_attentions:
-            # raise ValueError("WhisperFlashAttention2 attention does not support output_attentions")
             logger.warning_once("WhisperFlashAttention2 attention does not support output_attentions, "
                                 "manually calculating attention weights.")
 
@@ -105,7 +104,6 @@ class WhisperRoPEFlashAttn(WhisperFlashAttention2):
         key_states = self._reshape(self.k_proj(hidden_states), -1, bsz)
         query_states = self._reshape(self.q_proj(hidden_states), -1, bsz)
         if rotary_pos_emb is not None:
-            logger.warning_once("Using Rotary Position Embedding in WhisperRoPEFlashAttn. ")
             query_states, key_states = [apply_rotary_pos_emb(
                 i.transpose(1, 2),
                 rotary_pos_emb,