training_log.json

{
  "base_model_name": "Llama-2-13B-fp16-padded",
  "base_model_class": "LlamaForCausalLM",
  "base_loaded_in_4bit": true,
  "base_loaded_in_8bit": false,
  "projections": "q, v",
  "loss": 1.07,
  "learning_rate": 1.7080653142008908e-05,
  "epoch": 0.74,
  "current_steps": 9678,
  "train_runtime": 20262.2951,
  "train_samples_per_second": 12.817,
  "train_steps_per_second": 0.1,
  "total_flos": 4.119249176808653e+17,
  "train_loss": 1.1322209582423532
}