I’m seeing high variations in outputs between the Chat Completion API (gpt-3.5-turbo) and the Chat Playground. Correct me if I’m wrong, but they should be near identical with temperature set to 0.
The chat playground is set to:
- Temperature: 0
- Top P: 1
- Freq Penalty: 0
- Presence Penalty: 0
Does anyone know what I should pass through in the API to exactly match the Playground settings. There are some inconsistencies. For example, Presence Penalty range in Playground is 0 to 1. While in API documentation, the range is from -2 to 2, with a default of 0.
TLDR: Should my API presence penalty be 0 or -2 to match Playground setting of 0.