Meta AI کوئز

جب آپ کوئز دیتے ہیں تو جوابات اور وضاحتیں ظاہر ہوتی ہیں۔

1. Which downloadable model family belongs to Meta’s model ecosystem?
   • Llama
   • Claude
   • Gemini
   • Mistral
2. Before using a downloadable Llama release, what should a team read?
   • Only the family name on a blog post
   • That release’s model card, license, acceptable-use terms and supported configuration
   • A social post about a different checkpoint
   • Only the hosting provider’s marketing page
3. A benchmark on one Llama checkpoint is used to predict another size’s performance. What is the issue?
   • All sizes and modalities behave identically
   • Only hosted systems can be compared
   • Results do not automatically generalize across releases and configurations
   • Model cards cover only consumer assistants
4. A team downloads Llama weights to run a model locally. What else may it need?
   • Only a text editor
   • A social-media account
   • No other software or hardware
   • Compatible architecture code, tokenizer, configuration and suitable hardware
5. A quantized community build is being considered for deployment. What should the team do?
   • Evaluate that exact artifact directly
   • Assume it behaves identically to Meta’s original checkpoint
   • Use the base model card as proof of identical results
   • Skip tests because weights are downloadable
6. What does a downloadable checkpoint establish about reuse permissions?
   • That all uses are unrestricted
   • Nothing by itself; check the specific license and acceptable-use terms
   • That all training data are public
   • That the deployment has no risks
7. How can a hosted assistant differ from a downloaded checkpoint?
   • It is necessarily the same system with no added components
   • It never uses tools or policies
   • A product may add tools, policies and other components around a related model
   • It cannot be evaluated on representative tasks
8. Who must plan infrastructure, security, updates and monitoring for a local model deployment?
   • The weights themselves
   • The model’s training-data publishers
   • Only users who enter prompts
   • The organization operating the deployment