fix _forward_encoder_attention

vllm/v1/attention/backends/flash_attn.py

@@ -776,7 +776,7 @@ class FlashAttentionImpl(AttentionImpl):
                 # q_descale=layer._q_scale.expand(descale_shape),
                 # k_descale=layer._k_scale.expand(descale_shape),
                 # v_descale=layer._v_scale.expand(descale_shape),
-                is_prefix_cache=True,
+                is_prefix_cache=False,
             )
 
         return output