​

Chat Interface Overview

The chat interface provides a simple way to interact with your trained models.

​

Interface Layout

When you open the chat interface, you’ll see:

• Model selector - Choose which model to chat with
• Chat area - Conversation history and input
• Settings panel - Adjust generation parameters

​

Main Components

​

Model Selector

At the top, select your model:

• Enter a local path to your trained model
• Or use a Hugging Face model ID (e.g., meta-llama/Llama-3.2-1B)

​

Chat Area

The main conversation area:

• Message input - Type your messages here
• Send button - Submit your message
• Conversation history - See previous messages and responses

​

Settings Panel

Adjust how the model generates responses:

• Max tokens - Maximum response length
• Temperature - Creativity vs consistency (0.0 - 2.0)
• Top-p - Nucleus sampling threshold
• Top-k - Limit token choices
• Do Sample - Toggle sampling vs greedy decoding
• System Prompt - Set system instructions for the model

​

Basic Usage

1. Load a model - Enter model path and click Load
2. Type a message - Write in the input box
3. Send - Press Enter or click Send
4. Read response - Model generates a reply
5. Continue - Keep the conversation going

​

Tips

​

For Testing Fine-tuned Models

• Start with prompts similar to your training data
• Test edge cases and unusual inputs
• Compare responses to base model

​

For Finding Best Parameters

• Start with temperature 0.7 for balanced output
• Lower temperature (0.3) for more consistent answers
• Higher temperature (1.0+) for creative/varied responses

​

For Demos

• Prepare a few good example prompts
• Set parameters before the demo
• Clear conversation history between examples

​

Keyboard Shortcuts

​

Next Steps