Model configuration

Selecting and configuring the right AI model is crucial for your workflow's success. Lleverage helps you make informed decisions about which model to use and how to configure it for optimal results.

Choosing a model

The Model Advisor helps you select the right model by showing key performance metrics at a glance:

• Intelligence indicates the model's capability for complex reasoning and understanding. Higher intelligence scores mean better handling of complex tasks, nuanced understanding, and more sophisticated outputs.

• Speed shows how quickly the model responds. Higher speed ratings mean faster response times, crucial for real-time applications or when processing large amounts of data.

• Price is broken down into input and output costs per million tokens. This helps you understand the cost implications of your choices and optimize for your budget.

• Context size shows how much information the model can consider at once. Models with larger context can handle longer documents and maintain consistency across more complex conversations.

• Reliability indicates how consistently the model performs. Higher reliability scores mean more predictable and stable outputs.

Advanced settings

Once you've chosen a model, you can fine-tune its behavior through several parameters:

• Temperature controls the creativity and randomness in the model's responses. A higher temperature (closer to 1) makes responses more creative and varied, while lower values (closer to 0) make them more focused and deterministic.

• Top P (nucleus sampling) provides another way to control response variety. It determines how many different options the model considers when generating each part of its response.

• Frequency Penalty reduces repetition by making the model less likely to reuse the same information. Useful when you want diverse content without explicit repetition.

• Presence Penalty discourages the model from focusing too much on topics already mentioned. This helps ensure more balanced and comprehensive responses.

• Max Tokens sets the maximum length of the model's response. You can set this explicitly or use "Auto" to let Lleverage determine an appropriate limit based on your prompt.

• Seed allows you to get consistent results across multiple runs by using the same random seed. This is particularly useful when testing or when you need reproducible outputs.

Auto-tuning

Not sure which settings to use? The Auto Tune button can help. When clicked, it analyzes your prompt and use case to suggest optimal parameter settings. This is particularly useful when:

• You're just getting started with a new type of task

• You want to optimize your model's performance

• You're unsure about which parameters to adjust

After auto-tuning, you can still manually adjust any parameters to fine-tune the results further.

Best practices

Consider your use case when selecting a model. High intelligence isn't always necessary - sometimes a faster, more cost-effective model is the better choice.

Start with auto-tuned settings and adjust based on your results. Pay particular attention to temperature and max tokens, as these often have the most noticeable impact.

Monitor your costs and performance metrics to ensure your configuration remains optimal as your usage grows.