MetaMathQA/experiments/osf/llama-3.2-3B-rank128/training_params.json

{
  "optimizer_kwargs": {
    "lr": 5e-5
  }
}