checkpoint-100/trainer_state.json

{
  "best_metric": 0.47333332896232605,
  "best_model_checkpoint": "./zephyr/09-04-24-Weni-WeniGPT-Agents-Zephyr-1.0.17-KTO_Hyperparameter search, altering lora params for KTO task.-2_max_steps-145_batch_16_2024-04-09_ppid_9/checkpoint-100",
  "epoch": 0.684931506849315,
  "eval_steps": 50,
  "global_step": 100,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.14,
      "grad_norm": 3.690321445465088,
      "learning_rate": 0.00018,
      "loss": 0.4288,
      "step": 20,
      "train/kl": 4.702511787414551,
      "train/logps/chosen": -287.9634468129139,
      "train/logps/rejected": -294.00746579142015,
      "train/rewards/chosen": -0.4468543450563949,
      "train/rewards/margins": 1.449634002646083,
      "train/rewards/rejected": -1.8964883477024779
    },
    {
      "epoch": 0.27,
      "grad_norm": 9.52605017090491e-09,
      "learning_rate": 0.00015142857142857143,
      "loss": 0.4014,
      "step": 40,
      "train/kl": 11.584417343139648,
      "train/logps/chosen": -482.1775173611111,
      "train/logps/rejected": -527.2658025568181,
      "train/rewards/chosen": -20.685902913411457,
      "train/rewards/margins": 3.630714185310133,
      "train/rewards/rejected": -24.31661709872159
    },
    {
      "epoch": 0.34,
      "eval/kl": 0.0,
      "eval/logps/chosen": -1703.6540492957747,
      "eval/logps/rejected": -1555.5559731012659,
      "eval/rewards/chosen": -141.9996423855634,
      "eval/rewards/margins": -12.881070391892507,
      "eval/rewards/rejected": -129.11857199367088,
      "eval_loss": 0.47333332896232605,
      "eval_runtime": 139.1542,
      "eval_samples_per_second": 2.156,
      "eval_steps_per_second": 0.539,
      "step": 50
    },
    {
      "epoch": 0.41,
      "grad_norm": 0.0,
      "learning_rate": 0.00012285714285714287,
      "loss": 0.4781,
      "step": 60,
      "train/kl": 0.0,
      "train/logps/chosen": -1893.7743055555557,
      "train/logps/rejected": -1621.3242889221558,
      "train/rewards/chosen": -159.2800372753268,
      "train/rewards/margins": -25.875120546284876,
      "train/rewards/rejected": -133.40491672904193
    },
    {
      "epoch": 0.55,
      "grad_norm": 0.0,
      "learning_rate": 9.428571428571429e-05,
      "loss": 0.4813,
      "step": 80,
      "train/kl": 0.0,
      "train/logps/chosen": -1972.2258522727273,
      "train/logps/rejected": -1762.839984939759,
      "train/rewards/chosen": -167.9407721185065,
      "train/rewards/margins": -19.929229855705273,
      "train/rewards/rejected": -148.0115422628012
    },
    {
      "epoch": 0.68,
      "grad_norm": 0.0,
      "learning_rate": 6.571428571428571e-05,
      "loss": 0.4875,
      "step": 100,
      "train/kl": 0.0,
      "train/logps/chosen": -1872.451923076923,
      "train/logps/rejected": -1815.126524390244,
      "train/rewards/chosen": -158.9895958533654,
      "train/rewards/margins": -6.588621672725139,
      "train/rewards/rejected": -152.40097418064025
    },
    {
      "epoch": 0.68,
      "eval/kl": 0.0,
      "eval/logps/chosen": -1757.160761443662,
      "eval/logps/rejected": -1608.579509493671,
      "eval/rewards/chosen": -147.35032460387325,
      "eval/rewards/margins": -12.929376720170723,
      "eval/rewards/rejected": -134.42094788370252,
      "eval_loss": 0.47333332896232605,
      "eval_runtime": 138.9701,
      "eval_samples_per_second": 2.159,
      "eval_steps_per_second": 0.54,
      "step": 100
    }
  ],
  "logging_steps": 20,
  "max_steps": 145,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 1,
  "save_steps": 100,
  "total_flos": 0.0,
  "train_batch_size": 4,
  "trial_name": null,
  "trial_params": null
}