mxbai-embed-large-v1
Free.ai (self-hosted)
·
embeddings
·
~100 Ama-token ngalinye call
mxbai-embed-large-v1 yi an embedding model eyenziwe ngu mixedbread.ai. Onamandla kakhulu ku Semantic search, clustering, similarity.. I-self-hosted ku-Free.ai GPUs - isebenza ngokukhululekile ngaphezu kwe-token pool yakho yansuku zonke (100 tokens inombolo). Kukhishwa ngaphansi kwe-Apache 2.0 — ukusetshenziswa kokuthengiswa kwempahla kuvunyelwe ku-Free.ai.
Sebenzisa nge-API
I-REST API ehambisana ne-OpenAI. Dala isithonjana bese ubiza le modeli emaminithini.
curl -X POST https://api.free.ai/v1/image/generate/ \
-H "Authorization: Bearer sk-free-..." \
-H "Content-Type: application/json" \
-d '{"model":"mxbai-embed-large-v1","prompt":"your prompt here"}'Imibuzo ebuzwa kaningi
mxbai-embed-large-v1 iguqula umbhalo ube yi-vector eqinile (uhlu lwezibalo ezihambayo) ethatha uphawu. Sebenzisa ukusesha kwe-semantic, ukuqoqa, ukucebisa, ukukhishwa-ukuthuthukiswa kokukhiqizwa (RAG), kanye nanoma iyiphi imisebenzi lapho "iyikuphi le mbhalo ofana nalo mbhalo" ibalulekile.
Izilinganiso ezijwayelekile yi-384, 768, 1024, noma 1536 ngokuya ngemodeli. BGE-M3 ikhipha i-1024-dim; OpenAI Ada ikhipha i-1536. Umbuzo we-API ufaka isilinganiso ukuze i-vector DB yakho ikhethe isibalo esifanele.
Amamodeli amanje okungenisa (kufaka phakathi izinketho eziningi ku-Free.ai) aqeqeshiwe ku-100+ izilimi. Ukuthola kolimi oluphakathi kusebenza — khangela ngesiNgisi, hlanganisa amadokhumende ngesiSpanishi.
512 kuya ku-8,192 ama-token ngokuya ngemodeli. Izingeniso ezide zihlukaniswa — qoqa amadokhumende ade zibe ngamapharamitha ngaphambi kokufaka.
mxbai-embed-large-v1 isebenza kuma-GPU ethu athile futhi iyingxenye yezinto ezibiza kakhulu - mayelana nama-token angama-100 nganoma yisiphi isixhumanisi esikhishwa kusuka ku-pool yakho yamahhala yansuku zonke. $ 5 = ama-token angama-200K.
Yebo — thumela uhlu lwama-strings ku /v1/embeddings/ futhi mxbai-embed-large-v1 ibuyisela uhlu lwama-vectors ngokulandelana okufanayo. Ubukhulu be-batch kuya ku-2,048 ngesicelo ngasinye.
L2-ijwayelekile ngokuzenzakalela — i-cosine efana = i-dot product. Phatha `ijwayelekile = ayiqiniso` uma ufuna ama-vectors amnyama nge-metric yokuhamba okuhlukile.
Noma yini — Pinecone, Weaviate, Qdrant, Chroma, pgvector, FAISS, LanceDB. mxbai-embed-large-v1 ibuyisela i-JSON ejwayelekile; i-DB ayibonanga imodeli.
Yebo — POST ku /v1/embeddings/ ngemodeli="mxbai-embed-large-v1". Uhlobo lokuphendula oluhambisana ne-OpenAI, ngakho ama-client library akhona asebenza ngokungashintshwanga. /api/ unesixhumanisi esigcwele.
Amamodeli ahlala ahlala agcina umbhalo wakho kuma-GPUs ethu futhi asuse ngemuva kokubuyela kocingo. I-Premium idlula nge-DPA. Asiqeqeshi ngemingeniso yakho.
Sub-100ms for short text on self-hosted, 100–500ms on premium. Izingcingo ze-batch zilinganiselwa ngokuqondile — 1,000 chunks ziqediwe emaminithini angama-2–10.
Yebo — Free.ai inikeza ukusetshenziswa kokuthengiswa kwezinhlanganisela. Yenza ucwaningo lokukhishwa, amapayipi we-RAG, amasistimu wokucebisa ngaphandle kwe-royalty ye-vector ngayinye.