BGE-M3
Free.ai (self-hosted)
·
embeddings
·
~100 tokens per call
BGE-M3 waa an embedding model dhisay by BAAI. ugu xoog badan ee Text embeddings, semantic search. Is-hoosaysiinta Free.ai GPUs - bilaash ayaa ka socda ishaada maalinlaha ah (100 tokens wicitaan kasta). Soo saaray hoos MIT - isticmaalka ganacsi ee oggolaan Free.ai.
Ku isticmaal API
OpenAI-ku habboon REST API. abuuro furaha iyo wicidda noocan ah daqiiqado.
curl -X POST https://api.free.ai/v1/image/generate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"bge-m3","prompt":"your prompt here"}'Su'aalaha badanaa la isweydiiyo
BGE-M3 text u beddelaa vector ballaaran (listan oo ah flotates) oo macnaheedu yahay qabto. U isticmaal si loo hubiyo semantic, clustering, talooyin, soo kabashada-ku kordhay dhalasho (RAG), iyo hawl kasta oo ay "waa qoraalkan la mid ah qoraalka" arrimaha.
Miisaanka caadiga ah waa 384, 768, 1024, ama 1536 iyadoo ku xiran qaabka. BGE-M3 soo saartaa 1024-dim; OpenAI Ada soo saartaa 1536. jawaabta API waxaa ka mid ah miisaanka si aad vector DB dooran indekska saxda ah.
Modern embedding models (oo ay ku jiraan doorashooyinka ugu badan ee Free.ai) waxaa tababaray 100 + luqadood. Cross-luqadda soo kabashada shaqooyinka - raadinta Ingiriisi, la mid ah dukumiintiga in Spanish.
512 ilaa 8,192 calaamadaha ku xiran qaabka. Inputs dheer ayaa la gooyo - chunk dheer dukumiintiga in qodobbada ka hor intaan la embedding.
BGE-M3 ku socda GPUs our gaar ah oo waxaa ka mid ah qalabka ugu jaban - ku saabsan ~ 100 calaamadaha wicitaan kasta oo laga soo qaaday ka maalin kasta ee biyaha bilaashka ah. $5 = 200K calaamadaha.
Haa - POST liiska strings in /v1/embeddings / iyo BGE-M3 soo celin doonaa liiska vector in order isku mid ah. Batch size ilaa 2,048 codsiga kasta.
L2-normalized by default - cosine isku mid ah = dot wax soo saarka. Pass `normalize = false` haddii aad rabto vectors raw ee u dhexeeya metric kala duwan.
Wax kasta - Pinecone, Weaviate, Qdrant, Chroma, pgvector, FAISS, LanceDB. BGE-M3 ku soo laabtay caadiga ah JSON flotates; DB marnaba ma aragto qaabka.
Haa - POST in /v1/embeddings / oo leh model="BGE-M3". OpenAI-ka jawaabta qaabka la jaanqaada, sidaas darteed maktabadaha macaamiisha hadda jira shaqada aan la beddelin. /api/ waxay leedahay tilmaame buuxa.
Self-hosted qaababka ku hayn qoraalka ku GPUs our iyo ka saaro ka dib markii la soo noqnoqda. Premium la socdaan DPA. Waxaanan tababarka ku siin in aad inputs.
Sub-100ms qoraalka gaaban ee self-hosted, 100-500ms on premium. Baaqa baakadaha ayaa si toos ah u kordhiya - 1,000 chunks oo ku dhammaada 2-10 ilbiriqsi.
Haa - Free.ai siiya isticmaalka ganacsi ee embeddings. dhiso raadinta wax soo saarka, RAG pipelines, nidaamka talooyinka aan per-vector royalty.