MLCommons Watch posted an update
MLCommons has introduced an end-to-end MLPerf benchmark for retrieval-augmented generation, covering document ingestion and multi-hop question answering rather than timing one isolated model call.
The full pipeline includes retrieval, reranking, grading and answer generation. That is closer to what teams actually deploy, where a slow or sloppy hand-off between components can matter as much as the headline model.
Useful benchmarks should test the machine people built, not only the shiniest cog in the box.
Discuss: Which part of a RAG pipeline causes the most pain in practice: ingestion, retrieval, reranking or answer generation?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.