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

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

• Loss: 3.3908
• Accuracy: 0.4133

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
3.5989	1.0	18600	3.7674	0.3583
3.3772	2.0	37200	3.5890	0.3797
3.2519	3.0	55800	3.4638	0.3934
3.1739	4.0	74400	3.4033	0.3988
3.1143	5.0	93000	3.3692	0.4037
3.0706	6.0	111600	3.3750	0.4049
3.0371	7.0	130200	3.3584	0.4069
3.0052	8.0	148800	3.3419	0.4092
2.9778	9.0	167400	3.3557	0.4092
2.9507	10.0	186000	3.3506	0.4108
2.9315	11.0	204600	3.3575	0.4108
2.9052	12.0	223200	3.3518	0.4115
2.8856	13.0	241800	3.3580	0.4114
2.8675	14.0	260400	3.3460	0.4129
2.847	15.0	279000	3.3571	0.4130
2.8246	16.0	297600	3.3696	0.4134
2.8069	17.0	316200	3.3648	0.4141
2.7821	18.0	334800	3.3727	0.4136
2.7721	19.0	353400	3.3847	0.4136
2.7454	20.0	372000	3.3908	0.4133

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1