Ikibazo gihuza ibibazo · Hagati Urwego

Ikibazo cya Llama Model

Ongera usubiremo Llama ifunguye-uburemere bwasohotse, umurongo w'icyitegererezo, guhuza amabwiriza, gukora neza, hamwe nuburyo bushya bwa multimodal.

Inzira zijyanye no kuyoboraUmuryango wa Llama Model
Ikibazo 1 Bya 8

How does Meta describe access to Llama model weights, and what governs use?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. How does Meta describe access to Llama model weights, and what governs use?
    • Weights are publicly released for eligible versions, subject to each version’s license and use terms
    • Every version is public domain with unrestricted use
    • Only Meta employees can access any Llama weights
    • The weights are downloadable without any terms
  2. Amagambo ahinnye 'Llama' agereranya iki?
    • Kwiga kumurongo hamwe nubwubatsi bwa Multimodal
    • Icyitegererezo kinini cyururimi Meta AI
    • Isesengura ryumvikana hamwe nisesengura ryimashini
    • Porogaramu yo gucunga ururimi rwaho
  3. Nibihe Llama yarekuye yabanje kongeramo uruhushya rwemerera gukoresha ubucuruzi bwagutse?
    • Umwimerere Llama 1
    • Llama 405 gusa
    • Llama 2
    • Nta n'umwe muri bo wemera gukoresha ubucuruzi
  4. Ni ubuhe bwubatsi bushingiye kuri moderi ya Llama?
    • Ishusho yerekana ishusho
    • Icyemezo cyibiti
    • Umuyoboro usubiramo utitayeho
    • Impinduramatwara-decoder gusa ihanura ikimenyetso gikurikira
  5. Nigute Llama 'kuganira' cyangwa amabwiriza-yahinduwe kugirango akurikize abakoresha?
    • Binyuze mu kugenzura neza-gutunganya no gushimangira kwigira kubitekerezo byabantu
    • Mugukomeza ecran ya ecran
    • Mugusiba icyitegererezo cyibanze
    • Muguhindura ibisobanuro byose mukiratini mbere
  6. In what deployment setup can running Llama locally avoid sending prompts to a hosted inference service?
    • Whenever a model is downloaded, even when an app sends prompts to cloud tools
    • When inference runs on-device or on a private local server without a prompt-forwarding service
    • Only when the model is fine-tuned rather than used for inference
    • When the application stores prompts in a shared hosted API
  7. Which attention design did Meta adopt in Llama 3 8B and 70B to improve inference efficiency?
    • Residual vector quantization
    • Spatially-adaptive normalization
    • Grouped-query attention (GQA)
    • A recurrent state-space scan
  8. What inputs can Llama 3.2 Vision models accept?
    • Audio waveforms only
    • Video frames only, with no text
    • Sensor data without language or images
    • Text and images