Skill

GRPO

3 entries mention it. All skills →

Research

Studied how small models learn when given reasoning scaffolds, and compared the insides of base and fine-tuned models.

Notes & links

Development

Developed training tooling for Apple Silicon, with configurable tasks, evaluators, rewards, supervised modes and guided reasoning experiments.

Notes & links

Experiment

Kept training small models at home: guided-reasoning runs, quantisation and tooling to make runs repeatable.

Notes & links

From my own notes. Link to this entry