Using JAXforge
This guide walks you through the core components of the JAXforge application, showing you how to generate code, configure models, and save your work.
Core Components
The JAXforge application is designed to simulate and streamline the process of building a Transformer model. It is divided into several key components on the main page, each serving a specific purpose.
This is the AI-powered heart of the app. Instead of just a simple text box, it uses a multi-field input (Task Type, Latency, Database) to act as a 'prompt builder'. By providing these structured details, you help construct a high-quality, nuanced prompt for our Gemini-powered agent, resulting in more optimized and context-aware JAX/FLAX code.
This multi-tabbed card is your control center for model hyperparameters. You can adjust parameters manually, use AI to generate a configuration based on a task description, or load configurations you've previously saved.
This diagram shows the forward pass of a single Transformer Encoder Block. It illustrates the four key steps: Input (Tokenization & Embedding), Multi-Head Attention, Position-Wise Feed-Forward Network (FFN), and the residual 'Add & Norm' connections.
Your Keys, Your Control
We believe in transparency and user control. JAXforge is designed so that you are always in charge of your own intellectual property and resources.
Bring Your Own API Key
The "Settings" tab in the LLM Configurator allows you to enter your own Google AI (Gemini) API key. When you provide your own key, all AI-powered features (like code and config generation) are run through your personal account. We do not see or store your key; it is only kept in your browser for the duration of your session. This ensures that you have full control over API usage and costs.
Saving and Loading Configurations
One of the most powerful features of JAXforge is its integration with Firebase to manage your work.
Firebase Integration
To save your work, you must first sign in using the button in the header. The application uses Firebase's anonymous authentication, so you don't need to create an account to get started. Once signed in, a unique user profile is created for you in our Firestore database.
Live Updates
When you save a configuration (either manually or one generated by the AI), it's stored in a Firestore collection tied to your user ID. The 'Saved' tab in the LLM Configurator listens for real-time updates to this collection. This means any configurations you save will instantly appear in the list for you and any other authenticated users to see and use.