training_log.json

{
  "base_model_name": "NousResearch/Meta-Llama-3-8B-Instruct",
  "base_model_class": "LlamaForCausalLM",
  "base_loaded_in_4bit": true,
  "base_loaded_in_8bit": false,
  "projections": "gate, down, up, q, v, k, o",
  "loss": 2.1147,
  "grad_norm": 0.5170264840126038,
  "learning_rate": 1.6055383568469583e-09,
  "epoch": 0.9990897505916622,
  "current_steps": 5487,
  "train_runtime": 39825.0119,
  "train_samples_per_second": 0.138,
  "train_steps_per_second": 0.017,
  "total_flos": 5.117631701849211e+17,
  "train_loss": 2.262945574157092
}