learning_rate: 0.0001
train_batch_size: 256
eval_batch_size: 256
seed: 51
optimizer: Use adamw_torch_fused 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
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss
5.3946	1.0	423	4.2840
4.0574	2.0	846	4.0660
3.9206	3.0	1269	3.9986
3.8508	4.0	1692	3.9572
3.7992	5.0	2115	3.9278
3.7552	6.0	2538	3.9106
3.7154	7.0	2961	3.8962
3.6773	8.0	3384	3.8913
3.6399	9.0	3807	3.8849
3.6029	10.0	4230	3.8858
3.5662	11.0	4653	3.8891
3.5291	12.0	5076	3.8957
3.4924	13.0	5499	3.9041
3.457	14.0	5922	3.9158
3.4223	15.0	6345	3.9263
3.3905	16.0	6768	3.9369
3.3599	17.0	7191	3.9489
3.3346	18.0	7614	3.9587
3.3133	19.0	8037	3.9669
3.2976	20.0	8460	3.9721

fpadovani/candor_w_51

candor_w_51