I want embedding-backed passage rerankers for OpenDeepSearch.

I want embedding-backed passage rerankers for OpenDeepSearch.: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). calculate scores should embed the queries and documents, compute dot-product similarity with query embeddings @ doc embeddings.T , and support…

The task

`calculate_scores` should embed the queries and documents, compute dot-product similarity with `query_embeddings @ doc_embeddings.T`, and support `normalize="softmax"`, `normalize="scale"` by multiplying scores by 100, and `normalize="none"` with raw dot products. Passing an unknown normalization string should raise…

Part of FineEnvs/MiMo-V2.6-RL-harbor-code.