Core Concepts

How ModelFactory Works

ModelFactory integrates dataset access control and model fine-tuning into a seamless workflow, ensuring data security and ownership integrity. The platform’s primary processes include:

Dataset Request & Access

• Users submit dataset requests through OpenLedger’s repository.

• Providers approve or deny access based on established policies.

• Approved datasets are automatically linked to the user’s ModelFactory interface.

Model Selection & Configuration

• Users select from a wide range of LLMs (e.g., LLaMA, Mistral, DeepSeek).

• Hyperparameters like learning rate, batch size, and epochs are configured through the GUI.

Fine-Tuning Process

• The fine-tuning engine supports methods such as LoRA and QLoRA.

• Real-time dashboards provide training progress insights.

Evaluation & Deployment

• Built-in evaluation tools help analyze model performance.

• Exported models can be deployed in user applications or shared within the ecosystem.

Chat Interface

• Enables users to interact with fine-tuned models directly through the GUI.

• Supports real-time question-and-answer sessions or task-specific interactions.

RAG Attribution

• Provides retrieval-augmented generation (RAG) capabilities.

• When users ask a question, the model responds with generated outputs alongside sources of citations.

• Ensures proper attribution of information and enhances trust in generated outputs.