training-agents

A detailed tutorial on supervised fine-tuning (SFT) for training AI agents, built from scratch in pure PyTorch using Qwen3-0.6B, explaining the mechanics of next-token prediction and label masking.

A recap of a live stream where an AI agent (Codex) autonomously runs the entire SFT workflow to train a small Gemma 2B model to imitate a coding agent (pi). All artifacts and code are open-sourced.