InstructBLIP

InstructBLIP ni modeli ya lugha, kutambua picha na uchakataji wa data mseto iliyoundwa na Hong Kong University of Science and Technology (HKUST), Nanyang Technological University na Salesforce Research (Marekani, Hong Kong, Singapuri), iliyotolewa 11 Mei 2023.

Uzito wake uko wazi kwa utafiti na matumizi binafsi, lakini si kwa biashara.

Ina vigezo (parameters) bilioni 13.

Ilipotoka, ilivunja rekodi katika kazi yake.

Naweza kuiendesha mwenyewe?

Ndiyo, kwenye kompyuta mpakato au kifaa chenye uwezo mkubwa zaidi.

  1. SimuRAM ya GB 6–8
  2. Kompyuta mpakatoRAM ya GB 16
  3. Kompyuta yenye uwezo mkubwaGB 24 hadi 64 (kadi ya michoro au Mac yenye uwezo mkubwa)
  4. Sevazaidi ya GB 64

Inahitaji takriban GB 8.8 za kumbukumbu (RAM) ikibanwa hadi biti 4.

Programu kama Ollama au LM Studio zinapakua na kuendesha modeli kwenye kompyuta yako, bila intaneti.

Kumbuka: leseni yake hairuhusu matumizi ya biashara.

Modeli nyingine za Hong Kong University of Science and Technology (HKUST)

  1. MeissonicHong Kong University of Science and Technology (HKUST), National University of Singapore, Skywork AI, University of California (UC) Berkeley na Zhejiang University (ZJU)Mar 2025Kutengeneza picha
    Wazi1BSimu
  2. Oryx 34BNanyang Technological University, Tencent na Tsinghua UniversitySep 2024Data mseto, Utambuzi wa picha
    Wazi34BKompyuta yenye uwezo mkubwa
  3. Oryx 7BNanyang Technological University, Tencent na Tsinghua UniversitySep 2024Data mseto, Utambuzi wa picha
    Wazi7BKompyuta mpakato
  4. LLaVA-OV-72BByteDance, Chinese University of Hong Kong (CUHK), Hong Kong University of Science and Technology (HKUST) na Nanyang Technological UniversityAgo 2024Data mseto, Utambuzi wa picha
    Wazi72BKompyuta yenye uwezo mkubwa
  5. LLaVA-OV-7BByteDance, Chinese University of Hong Kong (CUHK), Hong Kong University of Science and Technology (HKUST) na Nanyang Technological UniversityAgo 2024Data mseto, Video
    Wazi7.6BKompyuta mpakato
  6. Grounding Dino LChinese University of Hong Kong (CUHK), Hong Kong University of Science and Technology (HKUST), International Digital Economy Academy, Microsoft Research, South China University of Technology na Tsinghua UniversityJul 2024Utambuzi wa picha
    Wazi341MSimu