training_log.json

{
  "base_model_name": "None",
  "base_model_class": "MistralForCausalLM",
  "base_loaded_in_4bit": false,
  "base_loaded_in_8bit": false,
  "projections": "gate, down, up, q, v, k, o",
  "loss": 0.118,
  "learning_rate": 1.195219123505976e-06,
  "epoch": 3.0,
  "current_steps": 8155,
  "train_runtime": 1400.5631,
  "train_samples_per_second": 11.659,
  "train_steps_per_second": 0.182,
  "total_flos": 1.8663500338220237e+17,
  "train_loss": 0.2664371884336659
}