agential-ai/agential

[Feature Request]: Diversifying prompts/benchmark ReAct few-shot example lengths

Closed this issue · 0 comments

Feature Description

A ReAct trajectory is:
Thought 1: ...
Action 1: ...
Observation 1: ...

The few-shot examples in prompts/benchmark that are of the ReAct style are somewhat uniform in trajectory length (the above trajectory is of length 1; each unit consists of a thought-action-observation triplet). Can we diversify the lengths a bit?

Reason

No response