AWD-LSTM - 3-layer LSTM (tied) + continuous cache pointer (WT2)

AWD-LSTM - 3-layer LSTM (tied) + continuous cache pointer (WT2) ni modeli ya lugha iliyoundwa na Salesforce Research (Marekani), iliyotolewa 7 Agosti 2017.

Haijatolewa kwa umma. Tunaijua kupitia ripoti au tangazo la walioiunda.

Ina vigezo (parameters) milioni 33.

Ni moja ya modeli zinazonukuliwa sana na watafiti.

Ilipotoka, ilivunja rekodi katika kazi yake.

Naweza kuiendesha mwenyewe?

Hapana.

Haiwezi kupakuliwa wala kutumika nje ya maabara iliyoiunda. Inaonekana hapa kwa sababu ilitangazwa au kuelezwa kwenye ripoti.

Modeli nyingine za Salesforce Research

  1. InstructBLIPHong Kong University of Science and Technology (HKUST), Nanyang Technological University na Salesforce ResearchMei 2023Data mseto, Lugha
    Wazi, si kwa biashara13BKompyuta mpakato
  2. BLIP-2 (Q-Former)Salesforce ResearchJan 2023Utambuzi wa picha, Lugha
    Wazi1.5BSimu
  3. Flan T5-XXL + BLIP-2Salesforce ResearchJan 2023Data mseto, Lugha
    Wazi12.1BKompyuta mpakato
  4. ProGen2-xlargeColumbia University, Johns Hopkins University na Salesforce ResearchJun 2022Biolojia
    Wazi6.4B
  5. 4 layer QRNN (h=2500)Salesforce ResearchMar 2018Lugha
    Haijatolewa151M
  6. QRNNSalesforce ResearchFeb 2018Lugha
    Haijatolewa135M