smolm-autoreg-bpe-counterfactual-babylm-old_union_new_regex_aanns_removal-1e-3

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

• Loss: 3.4290
• Accuracy: 0.4098

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: 0.001
• 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
3.603	1.0	18594	3.7800	0.3580
3.3857	2.0	37188	3.5982	0.3798
3.2557	3.0	55782	3.4973	0.3906
3.1798	4.0	74376	3.4102	0.3988
3.1213	5.0	92970	3.4141	0.4009
3.0811	6.0	111564	3.3950	0.4038
3.0439	7.0	130158	3.3901	0.4045
3.0119	8.0	148752	3.3961	0.4065
2.9874	9.0	167346	3.3729	0.4077
2.9569	10.0	185940	3.3747	0.4085
2.9339	11.0	204534	3.3791	0.4090
2.9152	12.0	223128	3.3890	0.4088
2.8931	13.0	241722	3.3884	0.4094
2.8717	14.0	260316	3.3770	0.4095
2.8513	15.0	278910	3.3836	0.4099
2.8363	16.0	297504	3.3995	0.4100
2.8109	17.0	316098	3.4114	0.4098
2.7931	18.0	334692	3.4218	0.4098
2.7786	19.0	353286	3.4200	0.4101
2.7601	20.0	371880	3.4290	0.4098

Framework versions

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