In Model Playground, which hyperparameters of an existing Salesforce-enabled foundational model can An Agentforce change?

Question

Options

• ATemperature, Frequency Penalty, Presence Penalty
• BTemperature, Top-k sampling, Presence Penalty
• CTemperature, Frequency Penalty, Output Tokens

• A
  88% (29)
• B
  6% (2)
• C
  6% (2)

Explanation

In Salesforce's Model Playground, the hyperparameters exposed for tuning on Salesforce-enabled foundational models are Temperature, Frequency Penalty, and Presence Penalty. Temperature controls randomness/creativity of outputs. Frequency Penalty reduces repetition of already-used tokens. Presence Penalty encourages the model to introduce new topics. Option B is incorrect because Top-k sampling is not an exposed parameter in this context. Option C is incorrect because Output Tokens (max token count) is a generation constraint, not a tunable hyperparameter in the Model Playground for these models.

No community discussion yet for this question.