checkpoint-4688/trainer_state.json
| 1 | { |
| 2 | "best_metric": 1.233594298362732, |
| 3 | "best_model_checkpoint": "fairface_age_image_detection/checkpoint-4688", |
| 4 | "epoch": 2.0, |
| 5 | "eval_steps": 500, |
| 6 | "global_step": 4688, |
| 7 | "is_hyper_param_search": false, |
| 8 | "is_local_process_zero": true, |
| 9 | "is_world_process_zero": true, |
| 10 | "log_history": [ |
| 11 | { |
| 12 | "epoch": 0.21331058020477817, |
| 13 | "grad_norm": 1.7164175510406494, |
| 14 | "learning_rate": 2.2574385510996123e-06, |
| 15 | "loss": 1.945, |
| 16 | "step": 500 |
| 17 | }, |
| 18 | { |
| 19 | "epoch": 0.42662116040955633, |
| 20 | "grad_norm": 1.979918360710144, |
| 21 | "learning_rate": 1.9879258300991807e-06, |
| 22 | "loss": 1.5803, |
| 23 | "step": 1000 |
| 24 | }, |
| 25 | { |
| 26 | "epoch": 0.6399317406143344, |
| 27 | "grad_norm": 2.0418295860290527, |
| 28 | "learning_rate": 1.7184131090987496e-06, |
| 29 | "loss": 1.4368, |
| 30 | "step": 1500 |
| 31 | }, |
| 32 | { |
| 33 | "epoch": 0.8532423208191127, |
| 34 | "grad_norm": 2.872443199157715, |
| 35 | "learning_rate": 1.4489003880983182e-06, |
| 36 | "loss": 1.3531, |
| 37 | "step": 2000 |
| 38 | }, |
| 39 | { |
| 40 | "epoch": 1.0, |
| 41 | "eval_accuracy": 0.5275, |
| 42 | "eval_loss": 1.3043227195739746, |
| 43 | "eval_model_preparation_time": 0.0035, |
| 44 | "eval_runtime": 894.9363, |
| 45 | "eval_samples_per_second": 11.174, |
| 46 | "eval_steps_per_second": 1.397, |
| 47 | "step": 2344 |
| 48 | }, |
| 49 | { |
| 50 | "epoch": 1.0665529010238908, |
| 51 | "grad_norm": 4.807504177093506, |
| 52 | "learning_rate": 1.179387667097887e-06, |
| 53 | "loss": 1.3097, |
| 54 | "step": 2500 |
| 55 | }, |
| 56 | { |
| 57 | "epoch": 1.2798634812286689, |
| 58 | "grad_norm": 6.204404830932617, |
| 59 | "learning_rate": 9.098749460974558e-07, |
| 60 | "loss": 1.2671, |
| 61 | "step": 3000 |
| 62 | }, |
| 63 | { |
| 64 | "epoch": 1.493174061433447, |
| 65 | "grad_norm": 6.795459270477295, |
| 66 | "learning_rate": 6.403622250970247e-07, |
| 67 | "loss": 1.2428, |
| 68 | "step": 3500 |
| 69 | }, |
| 70 | { |
| 71 | "epoch": 1.7064846416382253, |
| 72 | "grad_norm": 3.4942402839660645, |
| 73 | "learning_rate": 3.708495040965934e-07, |
| 74 | "loss": 1.2307, |
| 75 | "step": 4000 |
| 76 | }, |
| 77 | { |
| 78 | "epoch": 1.9197952218430034, |
| 79 | "grad_norm": 3.384183406829834, |
| 80 | "learning_rate": 1.0133678309616215e-07, |
| 81 | "loss": 1.2214, |
| 82 | "step": 4500 |
| 83 | }, |
| 84 | { |
| 85 | "epoch": 2.0, |
| 86 | "eval_accuracy": 0.5385, |
| 87 | "eval_loss": 1.233594298362732, |
| 88 | "eval_model_preparation_time": 0.0035, |
| 89 | "eval_runtime": 903.6391, |
| 90 | "eval_samples_per_second": 11.066, |
| 91 | "eval_steps_per_second": 1.383, |
| 92 | "step": 4688 |
| 93 | } |
| 94 | ], |
| 95 | "logging_steps": 500, |
| 96 | "max_steps": 4688, |
| 97 | "num_input_tokens_seen": 0, |
| 98 | "num_train_epochs": 2, |
| 99 | "save_steps": 500, |
| 100 | "stateful_callbacks": { |
| 101 | "TrainerControl": { |
| 102 | "args": { |
| 103 | "should_epoch_stop": false, |
| 104 | "should_evaluate": false, |
| 105 | "should_log": false, |
| 106 | "should_save": true, |
| 107 | "should_training_stop": true |
| 108 | }, |
| 109 | "attributes": {} |
| 110 | } |
| 111 | }, |
| 112 | "total_flos": 1.16245277051904e+19, |
| 113 | "train_batch_size": 32, |
| 114 | "trial_name": null, |
| 115 | "trial_params": null |
| 116 | } |
| 117 | |