Temperature determines how creative the model's responses are. Some guidelines:
0 - The model is deterministic, produces the same result given the same input every time, and is prone to repeating itself.
0.1 - 0.3 - The model is conservative and focused on the most likely response
0.7 - 0.9 - The model is creative and willing to take risks
1.0 - 2.0 - The model is very creative and may produce unexpected results
TopP restricts the model from randomly choosing unlikely tokens. It's effects are less obvious than temperature.
When the temperature is 0, this has no effect because the model only ever chooses the most likely value. Some guidelines:
0.2 - Model selects most likely words, resulting in safe and standard sentences.
0.9 - Model selects from a wider range of words, leaving to more diverse and interesting sentences.
Output tokens restricts the maximum amount of content the model will generate. 40 tokens are approximately 30 words.