illad_llama3/training_log.json

{
  "base_model_name": "Meta-Llama-3-8b",
  "base_model_class": "LlamaForCausalLM",
  "base_loaded_in_4bit": true,
  "base_loaded_in_8bit": false,
  "projections": "q, v",
  "loss": 3.1269,
  "grad_norm": 2.9809577465057373,
  "learning_rate": 1.5906680805938495e-09,
  "epoch": 3.0,
  "current_steps": 1985,
  "current_steps_adjusted": 1985,
  "epoch_adjusted": 3.0,
  "train_runtime": 3501.3176,
  "train_samples_per_second": 2.268,
  "train_steps_per_second": 0.567,
  "total_flos": 9.170665646063616e+16,
  "train_loss": 2.9956131780375648
}