Define
One task, a sample set and an agreed standard for the result.
OPEN-MODEL PROJECTS
A good model choice is one that works with your examples, your environment and your review process. The evaluation should make those trade-offs visible.
One task, a sample set and an agreed standard for the result.
The same examples, recorded settings and clear observations for each candidate.
An interface and runtime setup that the intended users can try.
Version information, test records and maintenance notes.
A short description of the task, representative material you are entitled to use, the intended users and any constraints on the operating environment. The first discussion can use a non-sensitive outline.
Not just successful answers. Include difficult inputs, missing information and cases that need human review. Compare the total effort around the answer as well as the response itself.
The model version and licence, where processing takes place, who has access, which integrations are included, and how the completed work will be assessed.