Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Isha Furan vs Lahaanshaha LLM ee Kediska Ganacsiga

Is barbar dhig Miisaanka furan iyo LLM-yada gaarka ah ee shuruudaha shatiga, xakamaynta xogta iyo wadarta qiimaha, oo baro sababta miisaanka furan aanu la mid ahayn isha furan.

Wadooyinka haga ee la xidhiidhaIsha Furan Vs Lahaanshaha LLM ee Ganacsiga
Su'aal 1 ee 8

What does the term open-weight most precisely mean?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. What does the term open-weight most precisely mean?
    • The model is free to use with no license terms
    • The model's training data is fully public
    • The model can only run on open-source hardware
    • The trained parameters can be downloaded, but training data and often code are not released
  2. Under Meta's Llama community license, which companies need a separate license from Meta?
    • Any company using the model commercially
    • Companies above 700 million monthly active users
    • Companies outside the United States
    • Companies with more than 100 employees
  3. Which license was Mistral 7B released under?
    • Apache 2.0
    • A custom community license with a user threshold
    • A proprietary API-only license
    • Wax shati ah haba yaraatee
  4. Which downside of proprietary models does the guide highlight?
    • They cannot be accessed through an API
    • They never offer enterprise support
    • The vendor can change prices, update behavior or retire model versions on its own schedule
    • They always use customer data for training
  5. How should a business confirm whether a proprietary provider trains on its API data?
    • Assume it never does
    • Assume it always does
    • Ask the model itself
    • Check the contract and terms rather than assuming
  6. Why does the guide recommend putting model calls behind a thin internal interface?
    • So switching providers means changing configuration rather than application code
    • To make models answer faster
    • To avoid paying for output tokens
    • Because licenses require it
  7. Why should you pin model versions where possible?
    • Pinned versions are always cheaper
    • Pinning removes all licensing obligations
    • An unannounced change in behavior can break prompts tuned for an earlier version
    • Pinning lets you download proprietary weights
  8. Which organization published an Open Source AI Definition in 2024?
    • Meta
    • The Open Source Initiative
    • The European Commission
    • NVIDIA