Properly Support Multi Tool Call Output

UKGovernmentBEIS / inspect_ai

Inspect: A framework for large language model evaluations

https://inspect.ai-safety-institute.org.uk/

MIT License

627 stars 118 forks source link

Properly Support Multi Tool Call Output #852

Closed dragonstyle closed 6 days ago

dragonstyle commented 1 week ago

This PR contains:

[ ] New features
[ ] Changes to dev-tools e.g. CI config / github tooling
[ ] Docs
[x] Bug fixes
[ ] Code refactor

If the model makes more than one tool call in a single message, we previously would capture on the last tool output and display that for all tool calls. This allows the message to carry more than one tool output for display.