smolm-autoreg-bpe-counterfactual-babylm-new_regex_aanns_removal-3e-4

This model was trained from scratch on the kanishka/counterfactual-babylm-new_regex_aanns_removal dataset. It achieves the following results on the evaluation set:

• Loss: 3.4031
• Accuracy: 0.4102

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 32000
• num_epochs: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.7373	1.0	18593	3.9120	0.3463
3.432	2.0	37186	3.6213	0.3756
3.296	3.0	55779	3.4757	0.3893
3.2046	4.0	74372	3.4592	0.3941
3.1486	5.0	92965	3.4210	0.3991
3.0995	6.0	111558	3.3986	0.4029
3.0612	7.0	130151	3.3767	0.4051
3.0315	8.0	148744	3.3788	0.4061
2.9984	9.0	167337	3.3602	0.4071
2.9731	10.0	185930	3.3580	0.4083
2.9506	11.0	204523	3.3490	0.4094
2.9303	12.0	223116	3.3534	0.4094
2.9062	13.0	241709	3.3573	0.4104
2.8838	14.0	260302	3.3740	0.4096
2.8665	15.0	278895	3.3801	0.4091
2.8447	16.0	297488	3.3746	0.4103
2.8233	17.0	316081	3.3850	0.4103
2.8093	18.0	334674	3.3949	0.4099
2.7876	19.0	353267	3.3955	0.4105
2.7743	20.0	371860	3.4031	0.4102

Framework versions

• Transformers 4.40.1
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.19.1