{ "best_metric": null, "best_model_checkpoint": null, "epoch": 12.5, "eval_steps": 500, "global_step": 5000, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "epoch": 0.125, "grad_norm": 0.044921875, "learning_rate": 0.00019800000000000002, "loss": 1.0694, "step": 50 }, { "epoch": 0.25, "grad_norm": 0.01007080078125, "learning_rate": 0.000196, "loss": 0.3955, "step": 100 }, { "epoch": 0.375, "grad_norm": 0.51171875, "learning_rate": 0.000194, "loss": 0.2071, "step": 150 }, { "epoch": 0.5, "grad_norm": 2.515625, "learning_rate": 0.000192, "loss": 0.5064, "step": 200 }, { "epoch": 0.625, "grad_norm": 0.0179443359375, "learning_rate": 0.00019, "loss": 0.4261, "step": 250 }, { "epoch": 0.75, "grad_norm": 4.625, "learning_rate": 0.000188, "loss": 0.3917, "step": 300 }, { "epoch": 0.875, "grad_norm": 7.53125, "learning_rate": 0.00018600000000000002, "loss": 0.5011, "step": 350 }, { "epoch": 1.0, "grad_norm": 0.00019168853759765625, "learning_rate": 0.00018400000000000003, "loss": 0.1436, "step": 400 }, { "epoch": 1.125, "grad_norm": 4.38690185546875e-05, "learning_rate": 0.000182, "loss": 0.2189, "step": 450 }, { "epoch": 1.25, "grad_norm": 0.0032196044921875, "learning_rate": 0.00018, "loss": 0.3036, "step": 500 }, { "epoch": 1.375, "grad_norm": 5.745887756347656e-05, "learning_rate": 0.00017800000000000002, "loss": 0.0851, "step": 550 }, { "epoch": 1.5, "grad_norm": 3.796875, "learning_rate": 0.00017600000000000002, "loss": 0.2121, "step": 600 }, { "epoch": 1.625, "grad_norm": 0.0037994384765625, "learning_rate": 0.000174, "loss": 0.0688, "step": 650 }, { "epoch": 1.75, "grad_norm": 0.000759124755859375, "learning_rate": 0.000172, "loss": 0.1541, "step": 700 }, { "epoch": 1.875, "grad_norm": 0.00113677978515625, "learning_rate": 0.00017, "loss": 0.1641, "step": 750 }, { "epoch": 2.0, "grad_norm": 5.602836608886719e-06, "learning_rate": 0.000168, "loss": 0.0284, "step": 800 }, { "epoch": 2.125, "grad_norm": 6.96875, "learning_rate": 0.000166, "loss": 0.0413, "step": 850 }, { "epoch": 2.25, "grad_norm": 6.496906280517578e-06, "learning_rate": 0.000164, "loss": 0.0338, "step": 900 }, { "epoch": 2.375, "grad_norm": 7.711350917816162e-07, "learning_rate": 0.000162, "loss": 0.0603, "step": 950 }, { "epoch": 2.5, "grad_norm": 7.62939453125e-06, "learning_rate": 0.00016, "loss": 0.0454, "step": 1000 }, { "epoch": 2.625, "grad_norm": 3.6656856536865234e-06, "learning_rate": 0.00015800000000000002, "loss": 0.0683, "step": 1050 }, { "epoch": 2.75, "grad_norm": 14.4375, "learning_rate": 0.00015600000000000002, "loss": 0.0216, "step": 1100 }, { "epoch": 2.875, "grad_norm": 4.947185516357422e-06, "learning_rate": 0.000154, "loss": 0.037, "step": 1150 }, { "epoch": 3.0, "grad_norm": 4.9591064453125e-05, "learning_rate": 0.000152, "loss": 0.033, "step": 1200 }, { "epoch": 3.125, "grad_norm": 7.003545761108398e-07, "learning_rate": 0.00015000000000000001, "loss": 0.0115, "step": 1250 }, { "epoch": 3.25, "grad_norm": 0.0009918212890625, "learning_rate": 0.000148, "loss": 0.0295, "step": 1300 }, { "epoch": 3.375, "grad_norm": 0.000263214111328125, "learning_rate": 0.000146, "loss": 0.0269, "step": 1350 }, { "epoch": 3.5, "grad_norm": 7.674098014831543e-07, "learning_rate": 0.000144, "loss": 0.0203, "step": 1400 }, { "epoch": 3.625, "grad_norm": 2.86102294921875e-06, "learning_rate": 0.000142, "loss": 0.0215, "step": 1450 }, { "epoch": 3.75, "grad_norm": 6.51925802230835e-07, "learning_rate": 0.00014, "loss": 0.0245, "step": 1500 }, { "epoch": 3.875, "grad_norm": 1.1548399925231934e-06, "learning_rate": 0.000138, "loss": 0.0204, "step": 1550 }, { "epoch": 4.0, "grad_norm": 2.637505531311035e-06, "learning_rate": 0.00013600000000000003, "loss": 0.0108, "step": 1600 }, { "epoch": 4.125, "grad_norm": 1.6689300537109375e-06, "learning_rate": 0.000134, "loss": 0.0068, "step": 1650 }, { "epoch": 4.25, "grad_norm": 2.3096799850463867e-07, "learning_rate": 0.000132, "loss": 0.0191, "step": 1700 }, { "epoch": 4.375, "grad_norm": 1.6689300537109375e-06, "learning_rate": 0.00013000000000000002, "loss": 0.0179, "step": 1750 }, { "epoch": 4.5, "grad_norm": 5.5730342864990234e-06, "learning_rate": 0.00012800000000000002, "loss": 0.0049, "step": 1800 }, { "epoch": 4.625, "grad_norm": 3.910064697265625e-05, "learning_rate": 0.000126, "loss": 0.028, "step": 1850 }, { "epoch": 4.75, "grad_norm": 3.609375, "learning_rate": 0.000124, "loss": 0.0099, "step": 1900 }, { "epoch": 4.875, "grad_norm": 0.00012111663818359375, "learning_rate": 0.000122, "loss": 0.0053, "step": 1950 }, { "epoch": 5.0, "grad_norm": 2.0116567611694336e-06, "learning_rate": 0.00012, "loss": 0.0147, "step": 2000 }, { "epoch": 5.125, "grad_norm": 0.31640625, "learning_rate": 0.000118, "loss": 0.0121, "step": 2050 }, { "epoch": 5.25, "grad_norm": 3.1739473342895508e-06, "learning_rate": 0.000116, "loss": 0.0057, "step": 2100 }, { "epoch": 5.375, "grad_norm": 2.3562461137771606e-07, "learning_rate": 0.00011399999999999999, "loss": 0.0116, "step": 2150 }, { "epoch": 5.5, "grad_norm": 2.473592758178711e-06, "learning_rate": 0.00011200000000000001, "loss": 0.0064, "step": 2200 }, { "epoch": 5.625, "grad_norm": 0.30859375, "learning_rate": 0.00011000000000000002, "loss": 0.005, "step": 2250 }, { "epoch": 5.75, "grad_norm": 2.816319465637207e-06, "learning_rate": 0.00010800000000000001, "loss": 0.0147, "step": 2300 }, { "epoch": 5.875, "grad_norm": 4.9114227294921875e-05, "learning_rate": 0.00010600000000000002, "loss": 0.0083, "step": 2350 }, { "epoch": 6.0, "grad_norm": 2.428889274597168e-06, "learning_rate": 0.00010400000000000001, "loss": 0.0071, "step": 2400 }, { "epoch": 6.125, "grad_norm": 5.602836608886719e-06, "learning_rate": 0.00010200000000000001, "loss": 0.0059, "step": 2450 }, { "epoch": 6.25, "grad_norm": 0.412109375, "learning_rate": 0.0001, "loss": 0.0103, "step": 2500 }, { "epoch": 6.375, "grad_norm": 0.6796875, "learning_rate": 9.8e-05, "loss": 0.0087, "step": 2550 }, { "epoch": 6.5, "grad_norm": 0.45703125, "learning_rate": 9.6e-05, "loss": 0.0038, "step": 2600 }, { "epoch": 6.625, "grad_norm": 2.7120113372802734e-06, "learning_rate": 9.4e-05, "loss": 0.0046, "step": 2650 }, { "epoch": 6.75, "grad_norm": 0.37890625, "learning_rate": 9.200000000000001e-05, "loss": 0.0066, "step": 2700 }, { "epoch": 6.875, "grad_norm": 4.5299530029296875e-05, "learning_rate": 9e-05, "loss": 0.0114, "step": 2750 }, { "epoch": 7.0, "grad_norm": 2.3096799850463867e-07, "learning_rate": 8.800000000000001e-05, "loss": 0.0028, "step": 2800 }, { "epoch": 7.125, "grad_norm": 2.4586915969848633e-06, "learning_rate": 8.6e-05, "loss": 0.0083, "step": 2850 }, { "epoch": 7.25, "grad_norm": 0.353515625, "learning_rate": 8.4e-05, "loss": 0.008, "step": 2900 }, { "epoch": 7.375, "grad_norm": 1.2665987014770508e-07, "learning_rate": 8.2e-05, "loss": 0.0042, "step": 2950 }, { "epoch": 7.5, "grad_norm": 1.2479722499847412e-07, "learning_rate": 8e-05, "loss": 0.0028, "step": 3000 }, { "epoch": 7.625, "grad_norm": 1.3317912817001343e-07, "learning_rate": 7.800000000000001e-05, "loss": 0.0077, "step": 3050 }, { "epoch": 7.75, "grad_norm": 2.384185791015625e-06, "learning_rate": 7.6e-05, "loss": 0.0047, "step": 3100 }, { "epoch": 7.875, "grad_norm": 1.0943040251731873e-07, "learning_rate": 7.4e-05, "loss": 0.002, "step": 3150 }, { "epoch": 8.0, "grad_norm": 1.3969838619232178e-07, "learning_rate": 7.2e-05, "loss": 0.0106, "step": 3200 }, { "epoch": 8.125, "grad_norm": 2.2649765014648438e-06, "learning_rate": 7e-05, "loss": 0.0037, "step": 3250 }, { "epoch": 8.25, "grad_norm": 2.8312206268310547e-06, "learning_rate": 6.800000000000001e-05, "loss": 0.0074, "step": 3300 }, { "epoch": 8.375, "grad_norm": 4.4405460357666016e-06, "learning_rate": 6.6e-05, "loss": 0.0046, "step": 3350 }, { "epoch": 8.5, "grad_norm": 2.518296241760254e-06, "learning_rate": 6.400000000000001e-05, "loss": 0.0052, "step": 3400 }, { "epoch": 8.625, "grad_norm": 2.8014183044433594e-06, "learning_rate": 6.2e-05, "loss": 0.0088, "step": 3450 }, { "epoch": 8.75, "grad_norm": 4.544854164123535e-07, "learning_rate": 6e-05, "loss": 0.0065, "step": 3500 }, { "epoch": 8.875, "grad_norm": 3.1851232051849365e-07, "learning_rate": 5.8e-05, "loss": 0.0079, "step": 3550 }, { "epoch": 9.0, "grad_norm": 5.304813385009766e-06, "learning_rate": 5.6000000000000006e-05, "loss": 0.0022, "step": 3600 }, { "epoch": 9.125, "grad_norm": 3.6954879760742188e-06, "learning_rate": 5.4000000000000005e-05, "loss": 0.0049, "step": 3650 }, { "epoch": 9.25, "grad_norm": 1.3131648302078247e-07, "learning_rate": 5.2000000000000004e-05, "loss": 0.0019, "step": 3700 }, { "epoch": 9.375, "grad_norm": 2.5480985641479492e-06, "learning_rate": 5e-05, "loss": 0.0036, "step": 3750 }, { "epoch": 9.5, "grad_norm": 0.439453125, "learning_rate": 4.8e-05, "loss": 0.0094, "step": 3800 }, { "epoch": 9.625, "grad_norm": 2.551823854446411e-07, "learning_rate": 4.600000000000001e-05, "loss": 0.0097, "step": 3850 }, { "epoch": 9.75, "grad_norm": 1.4808028936386108e-07, "learning_rate": 4.4000000000000006e-05, "loss": 0.0066, "step": 3900 }, { "epoch": 9.875, "grad_norm": 1.4156103134155273e-07, "learning_rate": 4.2e-05, "loss": 0.0041, "step": 3950 }, { "epoch": 10.0, "grad_norm": 2.682209014892578e-06, "learning_rate": 4e-05, "loss": 0.0046, "step": 4000 }, { "epoch": 10.125, "grad_norm": 1.2479722499847412e-07, "learning_rate": 3.8e-05, "loss": 0.0053, "step": 4050 }, { "epoch": 10.25, "grad_norm": 2.130866050720215e-06, "learning_rate": 3.6e-05, "loss": 0.0081, "step": 4100 }, { "epoch": 10.375, "grad_norm": 2.4028122425079346e-07, "learning_rate": 3.4000000000000007e-05, "loss": 0.0067, "step": 4150 }, { "epoch": 10.5, "grad_norm": 1.8477439880371094e-06, "learning_rate": 3.2000000000000005e-05, "loss": 0.0023, "step": 4200 }, { "epoch": 10.625, "grad_norm": 2.6226043701171875e-06, "learning_rate": 3e-05, "loss": 0.0052, "step": 4250 }, { "epoch": 10.75, "grad_norm": 2.7120113372802734e-06, "learning_rate": 2.8000000000000003e-05, "loss": 0.0042, "step": 4300 }, { "epoch": 10.875, "grad_norm": 0.447265625, "learning_rate": 2.6000000000000002e-05, "loss": 0.0056, "step": 4350 }, { "epoch": 11.0, "grad_norm": 2.905726432800293e-07, "learning_rate": 2.4e-05, "loss": 0.0058, "step": 4400 }, { "epoch": 11.125, "grad_norm": 2.9653310775756836e-06, "learning_rate": 2.2000000000000003e-05, "loss": 0.0022, "step": 4450 }, { "epoch": 11.25, "grad_norm": 3.0100345611572266e-06, "learning_rate": 2e-05, "loss": 0.0069, "step": 4500 }, { "epoch": 11.375, "grad_norm": 2.086162567138672e-07, "learning_rate": 1.8e-05, "loss": 0.0031, "step": 4550 }, { "epoch": 11.5, "grad_norm": 1.2889504432678223e-06, "learning_rate": 1.6000000000000003e-05, "loss": 0.0088, "step": 4600 }, { "epoch": 11.625, "grad_norm": 2.1606683731079102e-07, "learning_rate": 1.4000000000000001e-05, "loss": 0.0036, "step": 4650 }, { "epoch": 11.75, "grad_norm": 3.7066638469696045e-07, "learning_rate": 1.2e-05, "loss": 0.0063, "step": 4700 }, { "epoch": 11.875, "grad_norm": 0.44921875, "learning_rate": 1e-05, "loss": 0.0067, "step": 4750 }, { "epoch": 12.0, "grad_norm": 1.6205012798309326e-07, "learning_rate": 8.000000000000001e-06, "loss": 0.005, "step": 4800 }, { "epoch": 12.125, "grad_norm": 3.0100345611572266e-06, "learning_rate": 6e-06, "loss": 0.0058, "step": 4850 }, { "epoch": 12.25, "grad_norm": 1.1920928955078125e-07, "learning_rate": 4.000000000000001e-06, "loss": 0.0047, "step": 4900 }, { "epoch": 12.375, "grad_norm": 2.4139881134033203e-06, "learning_rate": 2.0000000000000003e-06, "loss": 0.0043, "step": 4950 }, { "epoch": 12.5, "grad_norm": 8.153915405273438e-05, "learning_rate": 0.0, "loss": 0.0032, "step": 5000 } ], "logging_steps": 50, "max_steps": 5000, "num_input_tokens_seen": 0, "num_train_epochs": 13, "save_steps": 1000, "stateful_callbacks": { "TrainerControl": { "args": { "should_epoch_stop": false, "should_evaluate": false, "should_log": false, "should_save": true, "should_training_stop": true }, "attributes": {} } }, "total_flos": 3.281807837380608e+16, "train_batch_size": 1, "trial_name": null, "trial_params": null }