llamahtml

LLamaHTML is a simple html file to communicate with a running llamacpp llama-server

I dislike the built-in html interface, so I made my own.

simply download the html file, and open in a web browser.

YOU MUST HAVE llama-server running

LlamaHTML Documentation

This web app provides a user interface for interacting with a local llama.cpp server, supporting streaming responses.

Features:

• Streaming Responses: Receives and renders model output in real-time as it's generated.
• Markdown Rendering: Displays model responses formatted with Markdown, including code highlighting.
• Code Block Actions: Copy and download code blocks from model output.
• Configurable Parameters: Adjust temperature, max tokens, top-p, top-k, and stop sequences.
• Theme Switching: Toggle between light and dark modes.
• Server URL Management: Specify the server URL and optionally save it to local storage.
• Clear Chat: Clears the conversation history.
• Abort Request: Cancel in-progress requests.

Setup:

1. Run llama.cpp server: Start your llama.cpp server locally (or remotely). Ensure the /completion endpoint is accessible. Use --host 0.0.0.0 if accessing the server from a different machine.
2. Enter Server URL: Provide the full URL of your server in the "Llama Server URL" field.
3. (Optional) Remember URL: Check "Remember URL" to store the server URL in your browser's local storage.

Usage:

1. Enter your message: Type your prompt in the input area.
2. Send: Click "Send" or press Ctrl+Enter (Cmd+Enter on macOS) to submit your prompt.
3. View Response: The model's response will stream into the conversation history area.
4. Configure (Optional): Use the "Configuration Options" to adjust parameters like temperature and max tokens.
5. Clear Chat: Click "Clear Chat" to reset the conversation.
6. Switch Theme: Click "Dark"/"Light" to toggle the theme.
7. Copy/Download Code: Click the "Copy" or "Download" buttons on code blocks to copy or download code.

Configuration Options:

• Temperature: Controls randomness (0 - ~2). Higher values = more random output. Default: 0.7
• Max Predict Tokens (n_predict): Maximum number of tokens to generate. -1 for infinite. Default: 2048.
• Top P (0-1): Nucleus sampling parameter. Default: 0.95
• Top K (integer): Top-k sampling parameter. Default: 40
• Stop Sequences: Comma-separated strings to stop generation. Default: USER:, \nUSER:, <|user|>

Troubleshooting:

• Connection Errors: Ensure the server is running and the URL is correct. Check the browser console for network errors.
• Invalid URL Format: The URL must be a valid HTTP or HTTPS URL (e.g., http://127.0.0.1:8080).
• Empty Message: You cannot send empty messages.

Dependencies:

• llama-server

Development:

The app is built with HTML, CSS, and JavaScript. The source code is available in the provided HTML file. You can modify and extend it as needed.