castorini-serve

You orchestrate a local HTTP stack by launching the Anserini RestServer and selectively enabling downstream Castorini services like rank_llm in a configurable, port-mapped layout. This supports cohesive retrieval, reranking, and generation workflows for local development and testing.

The best way to combine these workflows locally is using a flexible HTTP stack that maps ports to Anserini retrieval and optional Castorini services. You can compose retrieval-only, retrieval+rerank, nugget creation, and passage judging pipelines selectively.

Yes, an Anserini fatjar is required to serve the core retrieval functionality in the local HTTP stack. You must also provide optional models if you enable downstream services like rank_llm, ragnarok, and umbrela for reranking and judging.

Yes, you can test search-and-answer pipelines end-to-end using curl and jq pipelines. The stack provides reference materials and curated pipelines to guide real-world usage and quick demonstrations for local development and end-to-end testing.

You can compose retrieval-only, retrieval+rerank, generation, nugget creation, and passage judging pipelines in the local HTTP stack. Supported downstream Castorini services include rank_llm, ragnarok, nuggetizer, and umbrela for flexible service selection.

You should avoid a port-mapped HTTP stack if your local environment lacks the resources to run an Anserini fatjar and optional models simultaneously, or if your demos do not require combining retrieval with downstream reranking, generation, and judging services.