Documentation

What is halo forge?

halo forge is an RLVR (Reinforcement Learning from Verifiable Rewards) framework that uses compiler feedback as reward signals for iterative model refinement.

The Problem

The Approach

A compiler provides deterministic feedback — objective, reproducible results about code correctness.

Architecture

┌──────────┐    ┌──────────┐    ┌──────────┐    ┌───────────┐
│   Data   │ ─► │   SFT    │ ─► │   RAFT   │ ─► │ Benchmark │
└──────────┘    └──────────┘    └──────────┘    └───────────┘

1. Data — Gather training examples from public datasets or LLM generation
2. SFT — Supervised fine-tuning to establish baseline capability
3. RAFT — Iterative verification loop: generate → verify → filter → train
4. Benchmark — Evaluate with pass@k metrics

What to Expect

RAFT training typically shows:

Results vary significantly based on model, dataset, hardware, and domain. Run benchmarks to measure improvement on your specific use case.

Quick Navigation

Getting Started

• Quick Start — Get running in 30 minutes
• Toolbox Setup — Build the container environment
• Hardware Notes — Strix Halo configuration

Training Pipeline

• How to Train — Complete step-by-step guide (start here!)
• Full Pipeline — Complete training workflow
• Data Generation — Prepare training data
• SFT Training — Supervised fine-tuning
• RAFT Training — Reward-ranked fine-tuning
• Benchmarking — Evaluate with pass@k
• Production Runs — Production training commands

Verifiers

• Verifier Overview — Choose your verification strategy
• Compile Verifiers — GCC, Clang, MinGW, MSVC
• Test Verifiers — pytest, unittest
• Execution Verifiers — Test case verification
• Multi-Language — Auto-detect language
• Custom Verifiers — Build your own

Reference

• Command Index — Every command and flag
• Configuration — Config file reference
• Web UI — Dashboard for training and monitoring
• Windows Setup — MSVC build server
• Troubleshooting — Common issues

Background

• Theory & Research — Research foundations
• Graduated Rewards — Partial credit system
• Learning Rate Strategies — LR recommendations

Experimental

Features under active development and testing:

• Experimental Features — VLM, Audio, Reasoning, Agentic, Inference

Meta

• Changelog — Version history
• Contributing — How to contribute