LoRA Speedrun – a public wall-clock leaderboard for fine-tuning techniques
Summary
LoRA Speedrun presents a public, wall-clock leaderboard for fine-tuning LoRA on a frozen task (Qwen2.5-1.5B) with a single GPU and three verification runs. It describes current records, the verification workflow, and the community-driven brainstorm of ideas to improve parameter-efficient fine-tuning, underscoring reproducibility in ML benchmarking.