Skip to content

index.md

Latest commit

 

History

History
67 lines (45 loc) · 3.06 KB

index.md

File metadata and controls

67 lines (45 loc) · 3.06 KB
layout home
title Forgather: Democratizing Large Model Experimentation
description Alpha-stage ML framework bringing distributed pipeline parallelism to consumer GPUs. Seeking collaborators to help develop the future of accessible AI training.

Democratizing Large Model Experimentation

Forgather is an alpha-stage ML framework that aims to make large model training accessible to hobbyists and researchers with consumer hardware.

The Vision

Pipeline Parallelism for Consumer GPUs

Enable training of models larger than single GPU memory by distributing them across multiple consumer-grade cards using Torch Distributed Pipeline Parallelism Our goal is to make 7B+ parameter full model training accessible without enterprise hardware.

End Configuration Duplication

Eliminate the copy-paste cycle of ML experiments through a powerful template inheritance system. Specify only what changes between experiments, not entire configurations.

Framework-Independent Models

Generate standalone Python code that works without dependencies on Forgather itself. Your trained models remain portable and deployable anywhere.

Current Status: Alpha

Forgather is in active development with core functionality implemented:

  • Template inheritance system working
  • Pipeline parallelism implemented
  • Code generation pipeline functional
  • Multi-GPU distributed training
  • Performance optimization ongoing
  • Documentation and examples expanding

Seeking Collaborators

We're looking for contributors who share our vision of democratizing AI:

Researchers & Experimenters

Help test pipeline parallelism with different model architectures and share your findings.

ML Engineers

Contribute to performance optimization, memory efficiency, and distributed training improvements.

Documentation & Examples

Help create tutorials, guides, and example configurations for the community.

Core Development

Contribute to framework architecture, code generation, and template systems.

What We're Building

  • Accessible Training: 7B+ models on consumer setups
  • Template-Driven: Systematic experimentation without configuration chaos
  • Pipeline Parallelism: Much faster than FSDP on hardware lacking a fast interconnect
  • Framework Freedom: Generated models work independently
  • Research Focus: Built for exploration and comparison

Early Results

Initial testing on 4x RTX 4090 setup shows promising results for 7B parameter model training with various pipeline schedules. We're particularly excited about zero-bubble pipeline performance, though optimization work continues.

Explore the Code →{: .btn .btn-primary} Join Discussions →{: .btn .btn-outline} Report Issues →{: .btn .btn-outline}

Alpha software seeking alpha testers. Help us build the future of accessible AI training.