| { | |
| "best_global_step": 320, | |
| "best_metric": 0.9868217706680298, | |
| "best_model_checkpoint": "./multitask_model2/checkpoint-320", | |
| "epoch": 16.0, | |
| "eval_steps": 500, | |
| "global_step": 320, | |
| "is_hyper_param_search": false, | |
| "is_local_process_zero": true, | |
| "is_world_process_zero": true, | |
| "log_history": [ | |
| { | |
| "epoch": 1.0, | |
| "grad_norm": 2.824265241622925, | |
| "learning_rate": 1.9050000000000002e-05, | |
| "loss": 1.1436, | |
| "step": 20 | |
| }, | |
| { | |
| "epoch": 1.0, | |
| "eval_disagreement_accuracy": 0.539568345323741, | |
| "eval_disagreement_f1": 0.6049382716049383, | |
| "eval_loss": 1.0950782299041748, | |
| "eval_runtime": 1.0649, | |
| "eval_samples_per_second": 130.527, | |
| "eval_steps_per_second": 4.695, | |
| "eval_target_accuracy": 0.3237410071942446, | |
| "eval_target_f1": 0.23027834234730787, | |
| "step": 20 | |
| }, | |
| { | |
| "epoch": 2.0, | |
| "grad_norm": 2.4858803749084473, | |
| "learning_rate": 1.805e-05, | |
| "loss": 1.1033, | |
| "step": 40 | |
| }, | |
| { | |
| "epoch": 2.0, | |
| "eval_disagreement_accuracy": 0.5035971223021583, | |
| "eval_disagreement_f1": 0.5241379310344828, | |
| "eval_loss": 1.0678223371505737, | |
| "eval_runtime": 1.0741, | |
| "eval_samples_per_second": 129.413, | |
| "eval_steps_per_second": 4.655, | |
| "eval_target_accuracy": 0.4244604316546763, | |
| "eval_target_f1": 0.2658482564892094, | |
| "step": 40 | |
| }, | |
| { | |
| "epoch": 3.0, | |
| "grad_norm": 2.614800214767456, | |
| "learning_rate": 1.705e-05, | |
| "loss": 1.0708, | |
| "step": 60 | |
| }, | |
| { | |
| "epoch": 3.0, | |
| "eval_disagreement_accuracy": 0.48201438848920863, | |
| "eval_disagreement_f1": 0.4626865671641791, | |
| "eval_loss": 1.058626651763916, | |
| "eval_runtime": 1.0875, | |
| "eval_samples_per_second": 127.819, | |
| "eval_steps_per_second": 4.598, | |
| "eval_target_accuracy": 0.4028776978417266, | |
| "eval_target_f1": 0.3226851851851852, | |
| "step": 60 | |
| }, | |
| { | |
| "epoch": 4.0, | |
| "grad_norm": 2.5669617652893066, | |
| "learning_rate": 1.605e-05, | |
| "loss": 1.034, | |
| "step": 80 | |
| }, | |
| { | |
| "epoch": 4.0, | |
| "eval_disagreement_accuracy": 0.4892086330935252, | |
| "eval_disagreement_f1": 0.48175182481751827, | |
| "eval_loss": 1.039992094039917, | |
| "eval_runtime": 1.0925, | |
| "eval_samples_per_second": 127.234, | |
| "eval_steps_per_second": 4.577, | |
| "eval_target_accuracy": 0.45323741007194246, | |
| "eval_target_f1": 0.3739432367149758, | |
| "step": 80 | |
| }, | |
| { | |
| "epoch": 5.0, | |
| "grad_norm": 3.605031967163086, | |
| "learning_rate": 1.505e-05, | |
| "loss": 1.0189, | |
| "step": 100 | |
| }, | |
| { | |
| "epoch": 5.0, | |
| "eval_disagreement_accuracy": 0.48201438848920863, | |
| "eval_disagreement_f1": 0.4782608695652174, | |
| "eval_loss": 1.0271137952804565, | |
| "eval_runtime": 1.0502, | |
| "eval_samples_per_second": 132.36, | |
| "eval_steps_per_second": 4.761, | |
| "eval_target_accuracy": 0.4676258992805755, | |
| "eval_target_f1": 0.40272450554140693, | |
| "step": 100 | |
| }, | |
| { | |
| "epoch": 6.0, | |
| "grad_norm": 2.9558026790618896, | |
| "learning_rate": 1.4050000000000001e-05, | |
| "loss": 0.9716, | |
| "step": 120 | |
| }, | |
| { | |
| "epoch": 6.0, | |
| "eval_disagreement_accuracy": 0.49640287769784175, | |
| "eval_disagreement_f1": 0.5333333333333333, | |
| "eval_loss": 1.022552728652954, | |
| "eval_runtime": 1.0791, | |
| "eval_samples_per_second": 128.811, | |
| "eval_steps_per_second": 4.633, | |
| "eval_target_accuracy": 0.4244604316546763, | |
| "eval_target_f1": 0.38239258579165175, | |
| "step": 120 | |
| }, | |
| { | |
| "epoch": 7.0, | |
| "grad_norm": 3.551635265350342, | |
| "learning_rate": 1.305e-05, | |
| "loss": 0.9591, | |
| "step": 140 | |
| }, | |
| { | |
| "epoch": 7.0, | |
| "eval_disagreement_accuracy": 0.5035971223021583, | |
| "eval_disagreement_f1": 0.543046357615894, | |
| "eval_loss": 1.0097510814666748, | |
| "eval_runtime": 1.0876, | |
| "eval_samples_per_second": 127.8, | |
| "eval_steps_per_second": 4.597, | |
| "eval_target_accuracy": 0.43884892086330934, | |
| "eval_target_f1": 0.3822526173590003, | |
| "step": 140 | |
| }, | |
| { | |
| "epoch": 8.0, | |
| "grad_norm": 3.755647897720337, | |
| "learning_rate": 1.2050000000000002e-05, | |
| "loss": 0.9293, | |
| "step": 160 | |
| }, | |
| { | |
| "epoch": 8.0, | |
| "eval_disagreement_accuracy": 0.49640287769784175, | |
| "eval_disagreement_f1": 0.5333333333333333, | |
| "eval_loss": 1.0085021257400513, | |
| "eval_runtime": 1.3256, | |
| "eval_samples_per_second": 104.859, | |
| "eval_steps_per_second": 3.772, | |
| "eval_target_accuracy": 0.4316546762589928, | |
| "eval_target_f1": 0.3961486911322785, | |
| "step": 160 | |
| }, | |
| { | |
| "epoch": 9.0, | |
| "grad_norm": 3.108926296234131, | |
| "learning_rate": 1.1050000000000001e-05, | |
| "loss": 0.9066, | |
| "step": 180 | |
| }, | |
| { | |
| "epoch": 9.0, | |
| "eval_disagreement_accuracy": 0.5107913669064749, | |
| "eval_disagreement_f1": 0.5466666666666666, | |
| "eval_loss": 1.0002026557922363, | |
| "eval_runtime": 1.0775, | |
| "eval_samples_per_second": 128.998, | |
| "eval_steps_per_second": 4.64, | |
| "eval_target_accuracy": 0.45323741007194246, | |
| "eval_target_f1": 0.39100140056022403, | |
| "step": 180 | |
| }, | |
| { | |
| "epoch": 10.0, | |
| "grad_norm": 4.736630439758301, | |
| "learning_rate": 1.005e-05, | |
| "loss": 0.8903, | |
| "step": 200 | |
| }, | |
| { | |
| "epoch": 10.0, | |
| "eval_disagreement_accuracy": 0.539568345323741, | |
| "eval_disagreement_f1": 0.5844155844155844, | |
| "eval_loss": 1.0006366968154907, | |
| "eval_runtime": 1.0672, | |
| "eval_samples_per_second": 130.253, | |
| "eval_steps_per_second": 4.685, | |
| "eval_target_accuracy": 0.43884892086330934, | |
| "eval_target_f1": 0.4029991870962222, | |
| "step": 200 | |
| }, | |
| { | |
| "epoch": 11.0, | |
| "grad_norm": 4.827944278717041, | |
| "learning_rate": 9.050000000000001e-06, | |
| "loss": 0.8712, | |
| "step": 220 | |
| }, | |
| { | |
| "epoch": 11.0, | |
| "eval_disagreement_accuracy": 0.5323741007194245, | |
| "eval_disagreement_f1": 0.5806451612903226, | |
| "eval_loss": 0.9992592334747314, | |
| "eval_runtime": 1.0837, | |
| "eval_samples_per_second": 128.261, | |
| "eval_steps_per_second": 4.614, | |
| "eval_target_accuracy": 0.4460431654676259, | |
| "eval_target_f1": 0.41418171785130636, | |
| "step": 220 | |
| }, | |
| { | |
| "epoch": 12.0, | |
| "grad_norm": 4.064517498016357, | |
| "learning_rate": 8.050000000000001e-06, | |
| "loss": 0.8404, | |
| "step": 240 | |
| }, | |
| { | |
| "epoch": 12.0, | |
| "eval_disagreement_accuracy": 0.539568345323741, | |
| "eval_disagreement_f1": 0.5897435897435898, | |
| "eval_loss": 0.9953266382217407, | |
| "eval_runtime": 1.0689, | |
| "eval_samples_per_second": 130.037, | |
| "eval_steps_per_second": 4.678, | |
| "eval_target_accuracy": 0.4460431654676259, | |
| "eval_target_f1": 0.40262820512820513, | |
| "step": 240 | |
| }, | |
| { | |
| "epoch": 13.0, | |
| "grad_norm": 4.447686195373535, | |
| "learning_rate": 7.05e-06, | |
| "loss": 0.8457, | |
| "step": 260 | |
| }, | |
| { | |
| "epoch": 13.0, | |
| "eval_disagreement_accuracy": 0.539568345323741, | |
| "eval_disagreement_f1": 0.5897435897435898, | |
| "eval_loss": 0.992599606513977, | |
| "eval_runtime": 1.0579, | |
| "eval_samples_per_second": 131.393, | |
| "eval_steps_per_second": 4.726, | |
| "eval_target_accuracy": 0.4748201438848921, | |
| "eval_target_f1": 0.4418936499492189, | |
| "step": 260 | |
| }, | |
| { | |
| "epoch": 14.0, | |
| "grad_norm": 4.003559589385986, | |
| "learning_rate": 6.0500000000000005e-06, | |
| "loss": 0.838, | |
| "step": 280 | |
| }, | |
| { | |
| "epoch": 14.0, | |
| "eval_disagreement_accuracy": 0.539568345323741, | |
| "eval_disagreement_f1": 0.5897435897435898, | |
| "eval_loss": 0.9945887327194214, | |
| "eval_runtime": 1.0579, | |
| "eval_samples_per_second": 131.395, | |
| "eval_steps_per_second": 4.726, | |
| "eval_target_accuracy": 0.4748201438848921, | |
| "eval_target_f1": 0.448039914468995, | |
| "step": 280 | |
| }, | |
| { | |
| "epoch": 15.0, | |
| "grad_norm": 3.712287425994873, | |
| "learning_rate": 5.050000000000001e-06, | |
| "loss": 0.8099, | |
| "step": 300 | |
| }, | |
| { | |
| "epoch": 15.0, | |
| "eval_disagreement_accuracy": 0.5467625899280576, | |
| "eval_disagreement_f1": 0.5935483870967742, | |
| "eval_loss": 0.9911751747131348, | |
| "eval_runtime": 1.0757, | |
| "eval_samples_per_second": 129.216, | |
| "eval_steps_per_second": 4.648, | |
| "eval_target_accuracy": 0.4892086330935252, | |
| "eval_target_f1": 0.4611610088112851, | |
| "step": 300 | |
| }, | |
| { | |
| "epoch": 16.0, | |
| "grad_norm": 3.966646194458008, | |
| "learning_rate": 4.05e-06, | |
| "loss": 0.7843, | |
| "step": 320 | |
| }, | |
| { | |
| "epoch": 16.0, | |
| "eval_disagreement_accuracy": 0.539568345323741, | |
| "eval_disagreement_f1": 0.5949367088607594, | |
| "eval_loss": 0.9868217706680298, | |
| "eval_runtime": 1.0767, | |
| "eval_samples_per_second": 129.099, | |
| "eval_steps_per_second": 4.644, | |
| "eval_target_accuracy": 0.4748201438848921, | |
| "eval_target_f1": 0.4411908629018824, | |
| "step": 320 | |
| } | |
| ], | |
| "logging_steps": 500, | |
| "max_steps": 400, | |
| "num_input_tokens_seen": 0, | |
| "num_train_epochs": 20, | |
| "save_steps": 500, | |
| "stateful_callbacks": { | |
| "EarlyStoppingCallback": { | |
| "args": { | |
| "early_stopping_patience": 5, | |
| "early_stopping_threshold": 0.0 | |
| }, | |
| "attributes": { | |
| "early_stopping_patience_counter": 0 | |
| } | |
| }, | |
| "TrainerControl": { | |
| "args": { | |
| "should_epoch_stop": false, | |
| "should_evaluate": false, | |
| "should_log": false, | |
| "should_save": true, | |
| "should_training_stop": false | |
| }, | |
| "attributes": {} | |
| } | |
| }, | |
| "total_flos": 0.0, | |
| "train_batch_size": 32, | |
| "trial_name": null, | |
| "trial_params": null | |
| } | |