training_log.json

{
  "base_model_name": "llama3",
  "base_model_class": "LlamaForCausalLM",
  "base_loaded_in_4bit": false,
  "base_loaded_in_8bit": false,
  "projections": "q, v",
  "loss": 0.1027,
  "grad_norm": 0.08668982237577438,
  "learning_rate": 4.972375690607734e-06,
  "epoch": 2.968340824701974,
  "current_steps": 15163,
  "train_runtime": 13880.9358,
  "train_samples_per_second": 6.635,
  "train_steps_per_second": 0.013,
  "total_flos": 1.0810361631473664e+18,
  "train_loss": 0.3137563343256549
}