trainer_state.json

{
  "best_metric": null,
  "best_model_checkpoint": null,
  "epoch": 0.9874476987447699,
  "eval_steps": 500,
  "global_step": 59,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.02,
      "learning_rate": 1.6666666666666667e-09,
      "logits/chosen": -2.5598204135894775,
      "logits/rejected": -2.54148006439209,
      "logps/chosen": -231.18460083007812,
      "logps/pi_response": -153.6551971435547,
      "logps/ref_response": -153.6551971435547,
      "logps/rejected": -481.5467529296875,
      "loss": 0.6931,
      "rewards/accuracies": 0.0,
      "rewards/chosen": 0.0,
      "rewards/margins": 0.0,
      "rewards/rejected": 0.0,
      "step": 1
    },
    {
      "epoch": 0.17,
      "learning_rate": 9.860114570402054e-09,
      "logits/chosen": -2.579503297805786,
      "logits/rejected": -2.5002052783966064,
      "logps/chosen": -276.4703369140625,
      "logps/pi_response": -135.45098876953125,
      "logps/ref_response": -135.50714111328125,
      "logps/rejected": -483.57708740234375,
      "loss": 0.6932,
      "rewards/accuracies": 0.4756944477558136,
      "rewards/chosen": 0.0008863827679306269,
      "rewards/margins": 0.00028209862648509443,
      "rewards/rejected": 0.0006042842287570238,
      "step": 10
    },
    {
      "epoch": 0.33,
      "learning_rate": 8.374915007591053e-09,
      "logits/chosen": -2.543732166290283,
      "logits/rejected": -2.499950408935547,
      "logps/chosen": -286.89471435546875,
      "logps/pi_response": -147.18540954589844,
      "logps/ref_response": -147.16122436523438,
      "logps/rejected": -460.33807373046875,
      "loss": 0.6919,
      "rewards/accuracies": 0.6000000238418579,
      "rewards/chosen": -0.0019976082257926464,
      "rewards/margins": 0.0029969532042741776,
      "rewards/rejected": -0.004994561430066824,
      "step": 20
    },
    {
      "epoch": 0.5,
      "learning_rate": 5.738232820012407e-09,
      "logits/chosen": -2.52839732170105,
      "logits/rejected": -2.4818997383117676,
      "logps/chosen": -271.01190185546875,
      "logps/pi_response": -141.57553100585938,
      "logps/ref_response": -141.3872528076172,
      "logps/rejected": -487.7052307128906,
      "loss": 0.69,
      "rewards/accuracies": 0.6499999761581421,
      "rewards/chosen": -0.005116280633956194,
      "rewards/margins": 0.007777981460094452,
      "rewards/rejected": -0.012894262559711933,
      "step": 30
    },
    {
      "epoch": 0.67,
      "learning_rate": 2.8496739886173994e-09,
      "logits/chosen": -2.5488953590393066,
      "logits/rejected": -2.505760908126831,
      "logps/chosen": -280.8428955078125,
      "logps/pi_response": -146.02664184570312,
      "logps/ref_response": -145.77267456054688,
      "logps/rejected": -494.5501403808594,
      "loss": 0.6877,
      "rewards/accuracies": 0.6468750238418579,
      "rewards/chosen": -0.013240592554211617,
      "rewards/margins": 0.008154135197401047,
      "rewards/rejected": -0.021394727751612663,
      "step": 40
    },
    {
      "epoch": 0.84,
      "learning_rate": 6.947819411632222e-10,
      "logits/chosen": -2.562298536300659,
      "logits/rejected": -2.5050930976867676,
      "logps/chosen": -302.87677001953125,
      "logps/pi_response": -149.38168334960938,
      "logps/ref_response": -148.97946166992188,
      "logps/rejected": -509.5029296875,
      "loss": 0.6847,
      "rewards/accuracies": 0.699999988079071,
      "rewards/chosen": -0.017983367666602135,
      "rewards/margins": 0.016902435570955276,
      "rewards/rejected": -0.03488580137491226,
      "step": 50
    },
    {
      "epoch": 0.99,
      "step": 59,
      "total_flos": 0.0,
      "train_loss": 0.6886599670022221,
      "train_runtime": 3488.7106,
      "train_samples_per_second": 4.381,
      "train_steps_per_second": 0.017
    }
  ],
  "logging_steps": 10,
  "max_steps": 59,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 1,
  "save_steps": 100,
  "total_flos": 0.0,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}