non-leaf node如何表示

训练和检索的过程中，每一个item对应一个leaf-node且由对应的sparse embedding；那对于non-leaf node在进行prob计算的时候M的input怎么表示？

The issue mentions training, retrieval, leaf nodes, sparse embeddings, and probability calculation through M, but names no files, tests, or entry points. Start by locating the training and retrieval implementations and the code that computes probability with M. Done requires a documented, agreed representation for non-leaf node input.

Written by the indexing model from the issue text.