Suggested parameter combinations

Here are some suggested combination parameters for specific use cases that might be helpful.

Customer support

temperature: 0.2

top_p: 0.5

presence penalty: 0

frequency penalty: 0.6

In customer support, accurate and targeted responses are vital.

  • Lower temperature and top_p values ensure that the AI's responses are precise and on point.

  • Higher frequency penalty prevents the overuse of certain phrases.

Creative writing

temperature: 0.9

top_p: 0.9

presence Penalty: 0.6

frequency Penalty: 0

Creative writing requires imaginative and original content.

  • High temperature and top_p value encourages diverse and unpredictable outputs.

  • High presence penalty promotes novelty in the text

Academic writing

temperature: 0.2

top_p: 0.5

presence Penalty: 0.1

frequency Penalty: 0.6

Academic writing should be accurate, structured, and stay on the topic.

  • Low temperature and top_p values result in focused and reliable outputs.

  • Low presence penalty keeps the model on topic.

Social media post

temperature: 0.8

top_p: 0.9

presence Penalty: 0

frequency Penalty: 0.3

Social media posts should be engaging, casual, and often use recurring themes or hashtags.

  • High temperature and top_p value encourage creative and diverse outputs.

  • Low presence and frequency penalty values allow for the reuse of popular words, phrases, or topics that are common in social media trends.

Technical documentation

temperature: 0.3

top_p: 0.5

presence Penalty: 0.1

frequency Penalty: 0.6

Technical documentation needs to be clear, concise, and focused.

  • Lower temperature and top_p ensure deterministic and consistent output.

  • Low presence penalty helps the model stick to the specific technical topic.

Product description

temperature: 0.7

top_p: 0.9

presence Penalty: 0.1

frequency Penalty: 0.2

Product descriptions should be engaging, creative, and use repetitive phrases for emphasis.

  • Moderate temperature and high top_p value allow for a balance of creative and focused outputs.

  • Low presence penalty encourages sticking to the product's main features.

Last updated