ragflow icon indicating copy to clipboard operation
ragflow copied to clipboard

[Question]: Are the default values of the frequency_penalty template abnormal?

Open zstar1003 opened this issue 8 months ago • 1 comments

Self Checks

  • [x] I have searched for existing issues search for existing issues, including closed ones.
  • [x] I confirm that I am using English to submit this report (Language Policy).
  • [x] Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
  • [x] Please do not modify this template :) and fill in all the required fields.

Describe your problem

I noticed that in web/src/constants/knowledge.ts, this code defines three parameter style templates, and I have some doubts about the value of frequency_penalty. For creative writing, the model should be encouraged to use more diverse vocabulary rather than repeating words from the previous context, so frequency_penalty should ideally be set higher. However, I see that the current frequency_penalty for "Improvise" is set to 0.2—could this be a mistake?

export const settledModelVariableMap = {
  [ModelVariableType.Improvise]: {
    temperature: 0.9,
    top_p: 0.9,
    frequency_penalty: 0.2,
    presence_penalty: 0.4,
    max_tokens: 512,
  },
  [ModelVariableType.Precise]: {
    temperature: 0.1,
    top_p: 0.3,
    frequency_penalty: 0.7,
    presence_penalty: 0.4,
    max_tokens: 512,
  },
  [ModelVariableType.Balance]: {
    temperature: 0.5,
    top_p: 0.5,
    frequency_penalty: 0.7,
    presence_penalty: 0.4,
    max_tokens: 512,
  },
};

zstar1003 avatar Apr 03 '25 12:04 zstar1003

Hey there, @zstar1003! 👋 I'm here to help you with any bugs, questions, or becoming a contributor. Let's tackle this together!

The default value of the frequency_penalty for the "Improvise" model is set to 0.2, which is not abnormal. This value is within a typical range for frequency penalties, which usually range from 0 to 1. A frequency penalty of 0.2 is relatively low, suggesting a slight discouragement of repeated tokens. This aligns with the "Improvise" model's goal of generating more creative and varied outputs, as it allows for more freedom in word choice and structure [1].

To continue talking to Dosu, mention @dosu.


Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Bug Report | Other

dosubot[bot] avatar Apr 03 '25 12:04 dosubot[bot]