Exam AI-102 topic 7 question 15 discussion

-temperature -frequency penalty

Reduce repetition: Increase Frequency penalty to suppress repetition of words: Frequency Penalty modifies the probability of words seen frequently during training, making them less likely. NOT Presence Penalty: Presence Penalty modifies the probability of words in the input text, making them less likely to repeat in the output. Reduce Randomness of each response: Decrease the Temperature to make the answers more deterministic

Temperature is CORRECT because it controls the randomness of the responses generated by the model. Lowering the temperature reduces randomness, leading to more deterministic and focused responses, which is useful when you want to reduce the variability in conversations. Frequency penalty is CORRECT because it discourages the model from repeating the same words or phrases by applying a penalty to tokens that have already appeared frequently in the conversation. This helps in reducing the repetition of words.

1. Frequency penalty - This parameter reduces the model's likelihood of repeating the same word or phrase. Increasing the frequency penalty will help in reducing the repetition of words in the chatbot's responses. 2. Top P (or possibly Temperature) - To reduce the randomness of each response, you can adjust the Top P parameter, which controls the diversity of the tokens generated by narrowing down the token pool to the most likely ones. Lowering the Top P value (or Temperature) will result in more deterministic responses, thus reducing randomness.