TorchForge

TorchForge is a FastAPI-based engine designed to automate the generation, management, and execution of PyTorch-related scripts and models. It provides a structured framework for defining model registries, handling data feeds, and orchestrating complex ML workflows.

---

Current Status

We are currently in the Initial phase of rebuilding V2.

• Current Focus: Script and model generation logic.

Roadmap Overview

• Generation: Core script generation engine.

• [] Test Generation (after modifications)

• [] Design Complete Architecture and DB

• [] Documentation

• Inference: Support for generating inference-ready scripts.

• Execution: Running scripts via remote workers, training orchestration.

• Data Management: Saving results, versioning models/artifacts.

• Visualization: Pipeline visualization and result comparison dashboards.

• Frontend: Interactive UI for managing generation and monitoring runs.

• Infrastructure: Automatic VM provisioning and deployment pipelines.

• full plan

---

Getting Started

Prerequisites

• uv (Recommended for dependency management)
• Python 3.12+
• Docker (Optional, for services like Redis/PostgreSQL)

Setup

1. Clone and Install:

   uv sync

2. Configure Environment:

   cp .env.example .env
   # Edit .env with your local credentials

3. Initialize Database:

   uv run migrate

4. Run Development Server:

   uv run dev

---

Documentation

For detailed guides on development and architecture, please refer to the internal documentation:

• Scripting & Task Automation: How to use uv run shortcuts.
• Database Migrations: Handling schema changes with Alembic.
• Models & Data Schema: Overview of Feed, Script Generation, and Registry.
• Project Architecture: Core layout and engineering standards.

Tip

Run uv run docs to view the full interactive documentation site locally.