DeepSeek-V3.2

DeepSeek

Samo katalog Otvoreni utezi

Suverenost i hosting

Samo katalog

Ovaj je model u katalogu za referencu. Još nije hostan na LLM EU i nijedna vlastita krajnja točka ga ovdje ne poslužuje. Zahtjevi za njega idu izravno pružatelju, prema uvjetima tog pružatelja.

Sažetak

DeepSeek-V3.2 is a 685B mixture-of-experts model released under the MIT licence, with a configuration maximum of 163,840 tokens. Its main architectural change is DeepSeek Sparse Attention, which the vendor describes as reducing the computational cost of long-context inference. It is a text-only model and needs a multi-node deployment at full precision.

Specifikacije

Pružatelj
DeepSeek (CN)
Kontekstni prozor
164k
Parametri
685.36B
Modaliteti
text
Licenca
mit · licence
Otvoreni utezi
yes
Datum izlaska
01. pro 2025.
Zadaci
reasoning, coding, summarization, rag

Jezici

Jako: en, zh

Zadovoljavajuće: de, fr, es, it, pt, ru, ja, ko, ar

Jezici

Jako

  • en
  • zh

Zadovoljavajuće

  • de
  • fr
  • es
  • it
  • pt
  • ru
  • ja
  • ko
  • ar

Nije provjereno: pružatelj nije dokumentirao ovu tvrdnju. — a language is listed as strong only where a source claims it; the card links the evidence where one exists.

Kada ga koristiti

Prednosti

  • MIT licence
  • sparse attention aimed at lower long-context inference cost
  • 163,840-token configuration maximum

Ograničenja

  • 685B parameters: full-precision serving needs several nodes
  • vendor efficiency claims are not independently reproduced here
  • no evaluation has been run by LLM EU