gte-multilingual-reranker-base
Suveräänsus ja majutamine
See mudel on loetletud viitamiseks. Seda ei majutata veel LLM EU-s ega teeninda seda siin ükski esimese osapoole lõpp-punkt. Päringud suunatakse otse pakkujale, pakkuja enda tingimustel.
Kokkuvõte
gte-multilingual-reranker-base is a cross-encoder reranker for multilingual retrieval, with 305M parameters and support for inputs up to 8192 tokens. It is published under Apache-2.0 and covers 75 languages in its declared language list. In a retrieval pipeline it belongs after a first-stage vector search, not as a standalone embedding model.
Tehnilised andmed
- Pakkuja
- Alibaba Qwen (CN)
- Konteksti aken
- 8.2k
- Parameetrid
- 0.31B
- Modaalsused
- embedding
- Litsents
- apache-2.0 · licence
- Avatud kaalud
- yes
- Avaldamiskuupäev
- 20. juuli 2024
- Ülesanded
- rag, multilingual-support, private, cheap
Keeled
Tugev: en, de, fr, es, it, pt, nl, pl
Piisav: ru, uk, ar, zh, ja, ko, hi, tr, sv, da, fi, cs, ro, el
Keeled
Tugev
- en
- de
- fr
- es
- it
- pt
- nl
- pl
Piisav
- ru
- uk
- ar
- zh
- ja
- ko
- hi
- tr
- sv
- da
- fi
- cs
- ro
- el
Verifitseerimata: pakkuja ei ole seda väidet dokumenteerinud. — a language is listed as strong only where a source claims it; the card links the evidence where one exists.
Millal kasutada
Tugevused
- 8192-token input limit, long for a reranker
- 75 declared languages including all large EU languages
- small enough to run next to the embedding model
- Apache-2.0 licence
Piirangud
- a reranker only: it cannot generate text or embeddings for indexing
- the model card examples truncate inputs at 512 tokens unless configured otherwise
- no evaluation has been run by LLM EU