learning_rate: 0.001
train_batch_size: 96
eval_batch_size: 192
seed: 21
gradient_accumulation_steps: 8
total_train_batch_size: 768
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 20.0

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.8184	1.0	608	3.6983	0.3447
3.3613	2.0	1216	3.3208	0.3831
3.2083	3.0	1824	3.2031	0.3954
3.1128	4.0	2432	3.1444	0.4016
3.049	5.0	3040	3.1073	0.4056
2.9978	6.0	3648	3.0847	0.4084
2.9526	7.0	4256	3.0682	0.4107
2.9107	8.0	4864	3.0600	0.4118
2.8736	9.0	5472	3.0514	0.4131
2.8379	10.0	6080	3.0507	0.4137
2.8066	11.0	6688	3.0488	0.4146
2.7698	12.0	7296	3.0506	0.4148
2.7452	13.0	7904	3.0528	0.4153
2.7129	14.0	8512	3.0599	0.4153
2.6781	15.0	9120	3.0606	0.4156
2.6458	16.0	9728	3.0698	0.4154

qing-yao/binomial-babylm-base_seed-21_1e-3

binomial-babylm-base_seed-21_1e-3