Maswali yanayounganishwa na mwongozo · Kati Kiwango

Mifano ya Yi na Maswali 01.AI

Kagua asili ya familia ya mfano wa Yi, uzingatiaji wa lugha mbili, muundo wa dekoda pekee, anuwai za muktadha mrefu na mafunzo ya baada ya mafunzo yenye mwelekeo wa maelekezo.

Njia za mwongozo zinazohusianaMifano ya Yi na 01 AI
Swali 1 ya 8

Nani alianzisha 01.AI, kampuni nyuma ya mifano ya Yi?

Hakiki kila swali8 maswali

Majibu na maelezo huonekana unapojibu maswali.

  1. Nani alianzisha 01.AI, kampuni nyuma ya mifano ya Yi?
    • Kai-Fu Lee
    • Robin Li
    • Jack Ma
    • Andrew Ng
  2. Ni aina gani mbili za msingi za Yi zilizotolewa?
    • Yi-1 na Yi-2
    • Yi-6B na Yi-34B
    • Yi-100 na Yi-200
    • Yi-Mini na Yi-Max
  3. Je, aina za Yi zinajulikana kwa uwezo gani hasa?
    • Inazalisha picha pekee
    • Inaendesha bila GPU yoyote
    • Utendaji thabiti wa lugha mbili za Kichina-Kiingereza
    • Inatafsiri sauti inayozungumzwa pekee
  4. Ni nasaba gani ya msingi ya usanifu ambayo mifano ya Yi inafuata?
    • Muundo wa kawaida wa LSTM
    • Mfano wa kueneza
    • Muundo safi wa kusimba kama BERT
    • Transfoma ya avkodare pekee ya mtindo wa Llama
  5. Takriban dirisha la muktadha la anuwai za muktadha mrefu wa Yi lina urefu gani?
    • 200K tokeni
    • 4K ishara
    • 32K ishara
    • tokeni milioni 1
  6. What did 01.AI’s Yi technical report identify as an important driver of model performance?
    • Kuacha AI kabisa
    • Data quality resulting from data-engineering efforts
    • Kujenga tu chips za vifaa
    • Inaunganisha na OpenAI
  7. What does supervised fine-tuning plus reinforcement learning from human feedback aim to improve in Yi chat models?
    • Compressing retinal images
    • Changing the model into an encoder-only network
    • Following instructions in conversational use
    • Removing all text-generation ability
  8. What data strategy did 01.AI emphasize for the Yi models?
    • Training exclusively on synthetic audio
    • Avoiding all data filtering
    • Using only programming-language source code
    • Careful data quality and curation rather than relying only on sheer scale