checkpoint-100/trainer_state.json

{
  "best_metric": 0.16294628381729126,
  "best_model_checkpoint": "./Mistral/14-03-24-Weni-ZeroShot-3.4.7-Mistral-7b-DPO-1.0.0_ZeroShot DPO Training a improved dataset and best hyperparameters found so far-2_max_steps-144_batch_32_2024-03-14_ppid_9/checkpoint-100",
  "epoch": 2.0408163265306123,
  "eval_steps": 50,
  "global_step": 100,
  "is_hyper_param_search": false,
  "is_local_process_zero": true,
  "is_world_process_zero": true,
  "log_history": [
    {
      "epoch": 0.41,
      "grad_norm": 5.712127208709717,
      "learning_rate": 0.0001937984496124031,
      "logits/chosen": -0.9182177782058716,
      "logits/rejected": -0.8624125719070435,
      "logps/chosen": -20.338306427001953,
      "logps/rejected": -30.975902557373047,
      "loss": 0.436,
      "rewards/accuracies": 0.7093750238418579,
      "rewards/chosen": 0.03981679677963257,
      "rewards/margins": 1.3287527561187744,
      "rewards/rejected": -1.288935899734497,
      "step": 20
    },
    {
      "epoch": 0.82,
      "grad_norm": 2.6526589393615723,
      "learning_rate": 0.00016279069767441862,
      "logits/chosen": -0.16490581631660461,
      "logits/rejected": -0.056700706481933594,
      "logps/chosen": -17.591310501098633,
      "logps/rejected": -40.76393127441406,
      "loss": 0.3185,
      "rewards/accuracies": 0.856249988079071,
      "rewards/chosen": 0.4018515944480896,
      "rewards/margins": 2.6139047145843506,
      "rewards/rejected": -2.2120535373687744,
      "step": 40
    },
    {
      "epoch": 1.02,
      "eval_logits/chosen": -0.8333766460418701,
      "eval_logits/rejected": -0.7281509041786194,
      "eval_logps/chosen": -12.963512420654297,
      "eval_logps/rejected": -39.84817886352539,
      "eval_loss": 0.23576030135154724,
      "eval_rewards/accuracies": 0.8863636255264282,
      "eval_rewards/chosen": 0.7735208868980408,
      "eval_rewards/margins": 2.904942035675049,
      "eval_rewards/rejected": -2.1314210891723633,
      "eval_runtime": 78.4236,
      "eval_samples_per_second": 2.219,
      "eval_steps_per_second": 0.281,
      "step": 50
    },
    {
      "epoch": 1.22,
      "grad_norm": 3.620206356048584,
      "learning_rate": 0.0001317829457364341,
      "logits/chosen": -0.6694058179855347,
      "logits/rejected": -0.5974010229110718,
      "logps/chosen": -13.744880676269531,
      "logps/rejected": -37.68608093261719,
      "loss": 0.2533,
      "rewards/accuracies": 0.887499988079071,
      "rewards/chosen": 0.7727169990539551,
      "rewards/margins": 2.70760178565979,
      "rewards/rejected": -1.934884786605835,
      "step": 60
    },
    {
      "epoch": 1.63,
      "grad_norm": 1.9389252662658691,
      "learning_rate": 0.00010077519379844962,
      "logits/chosen": -0.7864009737968445,
      "logits/rejected": -0.7458208799362183,
      "logps/chosen": -14.465707778930664,
      "logps/rejected": -47.313682556152344,
      "loss": 0.1986,
      "rewards/accuracies": 0.910937488079071,
      "rewards/chosen": 0.6803622841835022,
      "rewards/margins": 3.599208354949951,
      "rewards/rejected": -2.9188458919525146,
      "step": 80
    },
    {
      "epoch": 2.04,
      "grad_norm": 3.7315940856933594,
      "learning_rate": 6.976744186046513e-05,
      "logits/chosen": -0.6323219537734985,
      "logits/rejected": -0.5668066740036011,
      "logps/chosen": -21.151926040649414,
      "logps/rejected": -48.24977493286133,
      "loss": 0.2257,
      "rewards/accuracies": 0.8984375,
      "rewards/chosen": -0.01782902702689171,
      "rewards/margins": 2.9440131187438965,
      "rewards/rejected": -2.9618420600891113,
      "step": 100
    },
    {
      "epoch": 2.04,
      "eval_logits/chosen": -0.6694537997245789,
      "eval_logits/rejected": -0.575288712978363,
      "eval_logps/chosen": -15.8816556930542,
      "eval_logps/rejected": -46.66548156738281,
      "eval_loss": 0.16294628381729126,
      "eval_rewards/accuracies": 0.8977272510528564,
      "eval_rewards/chosen": 0.48170650005340576,
      "eval_rewards/margins": 3.294857978820801,
      "eval_rewards/rejected": -2.8131513595581055,
      "eval_runtime": 78.4229,
      "eval_samples_per_second": 2.219,
      "eval_steps_per_second": 0.281,
      "step": 100
    }
  ],
  "logging_steps": 20,
  "max_steps": 144,
  "num_input_tokens_seen": 0,
  "num_train_epochs": 3,
  "save_steps": 100,
  "total_flos": 0.0,
  "train_batch_size": 8,
  "trial_name": null,
  "trial_params": null
}