BGE-M3
Free.ai (self-hosted)
·
embeddings
·
~100 Token kwa call
BGE-M3 bụ an embedding model e mepụtara site na BAAI. Ọrụ na Text embeddings, semantic search. Self-hosted na Free.ai GPUs - na-agba ọsọ n'efu megide ụbọchị gị token pool (100 tokens N'ime oku ọbụla). E wepụtara ya n'okpuru MIT — iji azụmahịa ekwenyela na Free.ai.
Jiri site na API
OpenAI-na-akpaghị aka REST API. Kewapụta kii nakwa kpọọ móòdù a n'ime sekọnd.
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"}'Ajụjụ ndị a jụrụkarị
BGE-M3 na-atụgharị ngwe n'ime vektor dị n'ime (ụdị ndesịta nke floats) nke na-echekwa ihe ọbụla. Jiri ya maka ọchụchọ semantikhi, n'ịrụkọta, n'enyemaka, n'ịrụpụta-amụbawanye (RAG), nakwa ọrụ ọbụla ebe "nke ngwe a dị ka nke ahụ" dị mkpa.
Ụdị nha bụ 384, 768, 1024, mọọbụ 1536 dabere na móòdù ahụ. BGE-M3 na-ewepụ 1024-dim; OpenAI Ada na-ewepụ 1536. Nnyeghachi API na-agụnye nha ka vektor DB gị na-ahọrọ indeksị ziri ezi.
Modern embedding models (na-agụnye nhọrọ ndị kasị na Free.ai) bụ n'ịgba ígwè na 100 + asụsụ. Cross-asụsụ retrieval ọrụ - nchọgharị na English, match faịlụ na Spanish.
512 ruo 8,192 token n'ihe oyiyi ahụ. Nhazi ndị dị ogologo a tụkwasịrị ha n'ime - wepụ ogologo dọkumenti ndị ahụ n'ime paragrafu tupú ịnyepụta.
BGE-M3 runs on our own GPUs and is among the cheapest tools — about ~100 tokens per call drawn from your daily free pool. $5 = 200K tokens.
Yabụ - POST ndesịta nke strings na /v1/embeddings/ na BGE-M3 na-eziga ndesịta nke vektor na usoro iheomume ahụ. Báà ụhara ruo 2,048 n'ọdịnihu.
L2-normalized site na difọ́ọ̀ltụ̀ - cosine similitudes = dot product. Pas̃ `normalize=false` ma ịchọrọ vektor raw maka n'ebe dị iche iche metric.
Ọbụla - Pinecone, Weaviate, Qdrant, Chroma, pgvector, FAISS, LanceDB. BGE-M3 na-ezigagharị JSON n'ime mmiri; DB anaghị ahụ móòdù ahụ.
Ya - POST ka /v1/embeddings/ na model="BGE-M3". OpenAI-dị n'otu na-egosipụta, yabụ na ndịna-eji ya na-arụ ọrụ na-enweghị mgbanwe. /api/ nwere ntụgharị zuru ezu.
Models na-echekwa onwe ha na-echekwa ngwe gị na GPUs anyị ma wepu ya mgbe oku na-abịa. Premium na-aga site na DPA. Anyị anaghị arụ ọrụ na init gị.
Sub-100ms maka ngwe n'onwe ya-n'ụlọ, 100-500ms na premium. Batị na-akpọkwa n'ụdị linearly - 1,000 chunks zuru ezu na 2-10 sekọnd.
Ee — Free.ai na-enye ikike iji embeddings n'ụzọ azụmahịa. Bipụta nchọgharị mmepe, RAG pipelines, usoroiheomume n'ihe nlereanya na-enweghị ikike nke vektor.