fix some bug

src/transformers/models/mixtral/modeling_mixtral.py

@@ -919,7 +919,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
             # states by `routing_weights` on the corresponding tokens (top-1 and top-2)
             current_state = hidden_states[None, top_x].reshape(-1, hidden_dim)
             current_hidden_states = expert_layer(current_state) * routing_weights[top_x, idx, None]
-            final_hidden_states[top_x].index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
+            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
         final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
         return final_hidden_states, router_logits
 
@@ -1315,7 +1315,11 @@ def _update_causal_mask(
         using_static_cache = isinstance(past_key_values, StaticCache)
         using_sliding_window_cache = isinstance(past_key_values, SlidingWindowCache)
 
-        if self.config._attn_implementation == "sdpa" and not (using_static_cache or using_sliding_window_cache):
+        if (
+            self.config._attn_implementation == "sdpa"
+            and not (using_static_cache or using_sliding_window_cache)
+            and not output_attentions
+        ):
             if AttentionMaskConverter._ignore_causal_mask_sdpa(
                 attention_mask,
                 inputs_embeds=input_tensor,