jRrSrSSKJr SSKJr SSKJr SSKJr Sr \"SS9S 5r SS jr SS jr SS jr g )aEmbedding helper for the ADE RAG playground. Reuses quber's local `EmbeddingService` (BAAI/bge-large-en-v1.5, 1024 dims), which runs on GPU when available and falls back to CPU. Both ingest (embed each chunk) and query (embed the question) go through the same model, so the vector spaces match. Local embeddings are used instead of a hosted API so the experiment has no external embedding dependency or quota. ) annotations) lru_cache)Listget_embedding_servicei)maxsizec[SS9$)Nlocal)providerr8/home/mande/repo/quber/src/quber/playground/embedding.py_servicers !' 22rcU(d/$[5RUSS9nUVs/sHo"R5PM sn$s snf)N ) batch_size)rembedtolist)textsarrrows r embed_textsrs=  *  5R  0C$' (CSJJLC (( (sAc [U/5S$)Nr)r)texts r embed_queryr%s v q !!rcpSSKJn UR5(aURX5$[ U5$)zEmbed a document's chunks: on the GPU worker when this host is set up to send work there, locally otherwise. The question at query time always embeds locally; only ingestion, thousands of chunks at once, is worth a worker.r)gpu)quber.playgroundr configuredrr)doc_keyrrs rembed_documentr")s. % ~~yy(( u rN)r List[str]returnList[List[float]])rstrr$z List[float])r!r&rr#r$r%)__doc__ __future__r functoolsrtypingrquber.db.embeddingsr EMBED_DIMrrrr"r rrr-sE #5   133)" r