计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
retrieval-evaluation【效率与协作】Run BEIR retrieval backends and metric evaluation for dense, sparse, lexical, API-backed, FAISS, and custom models.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Use this sub-skill when the task is to retrieve BEIR results, evaluate metrics, choose a first-stage backend, validate a custom retriever model, persist embeddings, or diagnose optional retrieval backends.
EvaluateRetrieval, backend classes, model protocols, metric names, and result/runfile APIs.python scripts/retrieval_smoke.pypython scripts/retrieval_smoke.py --output-dir beir-retrieval-smoke-outputpython scripts/inspect_optional_backends.py --jsonEvaluateRetrieval, dense exact search, dense FAISS search, sparse search, BM25/Elasticsearch lexical search, retrieval model wrappers, API-backed embedding wrappers, custom retrieval model protocols, metric evaluation, embedding persistence, and retrieval result export.GenericDataLoader, HFDataLoader, and BEIR JSONL/TSV validation to ../data-loading/SKILL.md.