Ikibazo gihuza ibibazo · Hagati Urwego

Gutegura Ibikoresho kubakozi ba LLM Ikibazo

Gerageza uburyo amazina yibikoresho nibisobanuro biyobora guhitamo icyitegererezo, kuki imigambi ikaze namakosa akorwa afite akamaro, nimpamvu ibikoresho byinshi bibabaza abakozi.

Inzira zijyanye no kuyoboraGutegura Ibikoresho Kubakozi ba LLM
Ikibazo 1 Bya 8

Ni ubuhe buryo bw'icyitegererezo bukoresha mu guhitamo uburyo bwo guhamagara igikoresho?

Reba buri kibazo8 ibibazo

Ibisubizo nibisobanuro bigaragara mugihe ufata ikibazo.

  1. Ni ubuhe buryo bw'icyitegererezo bukoresha mu guhitamo uburyo bwo guhamagara igikoresho?
    • The tool's source code
    • The name, description, schema and returned text
    • The server's latency history
    • The programming language the tool is written in
  2. Kuki umukozi ashobora gukoresha prefixes nka github_create_issue na jira_create_issue?
    • Prefixes make tools run faster
    • APIs require prefixes
    • They keep similar tools from different services apart
    • They reduce the size of JSON Schema files
  3. Ni ubuhe buryo bwo gushushanya bushobora gutanga ibitekerezo byukuri?
    • An enum listing the allowed values
    • A free-text string called 'value'
    • An untyped field with no description
    • A parameter named 'x'
  4. Niki gikora ubutumwa bwiza bwibikoresho?
    • A numeric error code only
    • A generic 'something went wrong'
    • Nothing, so no tokens are spent
    • It explains what went wrong and suggests what to try next
  5. Why can adding many tools reduce agent performance?
    • Models can only call one tool per conversation
    • Each definition takes up context, and overlapping tools make selection harder
    • Tools stop working once there are more than five
    • Tool count has no effect
  6. Compared with mirroring every API endpoint one-to-one, what does the guide recommend?
    • Fewer, higher-level tools shaped around real tasks
    • Exposing every endpoint so the agent has full coverage
    • Using only one tool for everything
    • Letting the model write its own HTTP requests
  7. Why should a tool paginate or truncate its output by default?
    • To hide data from users
    • Because models cannot read numbers
    • To avoid filling the context window with low-value data
    • Because JSON Schema requires it
  8. Why make tools idempotent or give them an idempotency key?
    • It makes descriptions shorter
    • It lets the model skip schemas
    • It speeds up model inference
    • A retried call will not create duplicate records