MetaMathQA/default_training_params.json

{
  "model_id": "meta-llama/Llama-3.2-3B",
  "dtype": "bfloat16",
  "max_seq_length": 768,
  "batch_size": 4,
  "batch_size_eval": 50,
  "max_steps": 5000,
  "eval_steps": 250,
  "compile": false,
  "seed": 0,
  "grad_norm_clip": 1.0,
  "optimizer_type": "AdamW",
  "optimizer_kwargs": {
    "lr": 1e-4,
    "weight_decay": 0.1
  },
  "lr_scheduler": "cosine",
  "use_amp": false,
  "autocast_adapter_dtype": true,
  "attn_implementation": null,
  "generation_kwargs": {
    "max_length": 800,
    "max_new_tokens": 300
  },
  "query_template": "Question: {query} Think step by step.\nAnswer:"
}