Skip to main content
Version: 0.0.38

Parameters

The Parameters tab within Custom Models allows users to fine-tune model behavior through various adjustable settings. Given the extensive support for diverse providers by Lakehousecat and the continuous evolution of models, parameter settings should be approached with caution. Their efficacy can vary depending on the model and its use case.

Introduction to Parameters​

Parameter adjustments are primarily intended for those exploring model behaviors or conducting in-depth analysis. While model parameters can greatly influence interaction outcomes, they may not be applicable across all settings or provider models. Testing thoroughly before relying on any adjustments for productive use is critical.

Key Parameters​

Below are some of the main parameters available for configuration:

Temperature​

  • Impact: Controls the randomness of model responses. Higher values lead to more diverse outputs, while lower values result in more deterministic replies.
  • Use Case: Useful for exploratory interactions or generating creative content.

Top K​

  • Definition: Limits the number of tokens considered by the model when generating a response.
  • Functionality: Applying this parameter can refine response diversity by selectively focusing on a subset of possible tokens.

Top P (Nucleus Sampling)​

  • Concept: It's an alternative to Top K, utilizing cumulative probability to determine token selection.
  • Application: Facilitates more nuanced output by balancing response creativity and coherence.

Stop Sequence​

  • Purpose: Defines specific sequences prompting the cessation of model output.
  • Utility: Useful for specifying the completion of generated content or controlling the length of interactions.

Best Practices​

  • Experiment With Caution: Adjust parameters only when you have a clear understanding of their impact or are actively experimenting.
  • Maintaining Default Settings: If uncertain, retain default parameters to ensure stable operation unless exploring new interaction possibilities.
  • Thorough Testing: Conduct comprehensive tests to verify the efficacy of parameter adjustments in harmony with model functions.

Conclusion​

Custom Models in Lakehousecat offer parameter configuration options that enhance flexibility in data interactions. By prudently experimenting with these settings, users can tailor models to suit unique requirements, though caution is advised to maintain system integrity and effective outputs.