Preserve RAG provenance through chunking and retrieval
Preserve RAG provenance through chunking and retrieval: a task in MiMo-V2.6-RL-harbor-code: MiMo-V2.6-RL Code (Harbor) (Harbor dataset). RAG answers can cite a source only if the source survives the entire retrieval path. At the moment a loader can produce a TextArtifact with a Reference and…
The task
RAG answers can cite a source only if the source survives the entire retrieval path. At the moment a loader can produce a `TextArtifact` with a `Reference` and useful source metadata, but chunking turns it into anonymous artifacts. Equal excerpt text from different sources is then collapsed as one result, so the…
Part of FineEnvs/MiMo-V2.6-RL-harbor-code.