LoRA Speedrun – a public wall-clock leaderboard for fine-tuning techniques

LoRA Speedrun is a public leaderboard that benchmarks the speed of fine-tuning techniques for large language models on fixed hardware (L40S GPU) and tasks, requiring independent verification of all records with fresh seeds. The competition features two tracks: fine-tuning Qwen2.5-1.5B on GSM8K math problems (current record: 6m 05s) and SmolLM2-1.7B on SQuAD reading comprehension (current record: 11m 08s), with anyone able to compete and verify results using free Modal compute credits.