Update modeling_Llamoe.py

modeling_Llamoe.py

@@ -660,9 +660,8 @@ class LlamoeSdpaAttention(LlamoeAttention):
         key_states = repeat_kv(key_states, self.num_key_value_groups)
         value_states = repeat_kv(value_states, self.num_key_value_groups)
 
-        causal_mask = attention_mask
-        if attention_mask is not None and cache_position is not None:
-            causal_mask = torch.tril(torch.ones((bsz, q_len, q_len), device=query_states.device))
+        
+        causal_mask = torch.tril(torch.ones((bsz, q_len, q_len), device=query_states.device))
 
         # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
         # Reference: https://github.com/pytorch/pytorch/issues/112577.