smolm-autoreg-bpe-counterfactual_babylm_aann_dtanns-1e-4

This model was trained from scratch on the kanishka/counterfactual_babylm_aann_dtanns dataset. It achieves the following results on the evaluation set:

• Loss: 3.4122
• Accuracy: 0.4058

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: 0.0001
• 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
4.0526	1.0	18595	4.2653	0.3097
3.5753	2.0	37190	3.7488	0.3616
3.3983	3.0	55785	3.5949	0.3788
3.2911	4.0	74380	3.5416	0.3863
3.2282	5.0	92975	3.4673	0.3924
3.1707	6.0	111570	3.4627	0.3948
3.1335	7.0	130165	3.4265	0.3989
3.1003	8.0	148760	3.4135	0.4001
3.0686	9.0	167355	3.4013	0.4009
3.0407	10.0	185950	3.3952	0.4028
3.0138	11.0	204545	3.3950	0.4027
2.9937	12.0	223140	3.3930	0.4036
2.9758	13.0	241735	3.4018	0.4041
2.9478	14.0	260330	3.4024	0.4045
2.9311	15.0	278925	3.4006	0.4050
2.9098	16.0	297520	3.4037	0.4050
2.8959	17.0	316115	3.3980	0.4054
2.8768	18.0	334710	3.4074	0.4058
2.8627	19.0	353305	3.4091	0.4058
2.8505	20.0	371900	3.4122	0.4058

Framework versions

• Transformers 4.38.0
• Pytorch 2.3.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.2