MerchantryTidbits

evals-testing / library

Weave

Symptoms: opaque agent failures; no prompt version history.

Use it when

  • trace multi-step LLM apps with W&B Weave
  • compare model routes and prompt versions with traces

What it solves

no-regression-signalmulti-model-routing

Not the fit when

  • not a model host

Install

pip install weave

Invoke

Decorate LLM calls with weave.op to trace prompts and scores in W&B Weave UI.

Alternatives

No reviewed alternatives recorded yet.