关于lshne中的single_view_loss的含义的疑问

请教下：感觉single_view_loss这部分loss贡献度很少，信息都被multi_view attention的loss贡献了，single_view_loss的意义是类似transformer中的纠偏，防止学的太差？

如果是这样的是不是对attention的信息量很不自信^_^ , 是不是需要加个参数，p q 去控制下loss 的weight更好。

No files, tests, or entry points are identified. Start by locating the lshne implementation and tracing how single_view_loss and multi_view attention loss are computed and combined; clarify the intended weighting behavior before deciding whether configurable p and q parameters are needed.

Written by the indexing model from the issue text.