training_log.json

{
  "base_model_name": "llama-saiga",
  "base_model_class": "LlamaForCausalLM",
  "base_loaded_in_4bit": false,
  "base_loaded_in_8bit": true,
  "loss": 2.4306,
  "learning_rate": 0.0,
  "epoch": 3.0,
  "current_steps": 4319,
  "train_runtime": 3963.6329,
  "train_samples_per_second": 4.36,
  "train_steps_per_second": 0.034,
  "total_flos": 1.7670854947110912e+17,
  "train_loss": 2.5247063248245802
}