smolm-autoreg-bpe-counterfactual_babylm_indef_articles_with_pl_nouns_removal_new-seed_1024-1e-3

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

• Loss: 3.4015
• Accuracy: 0.4107

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: 1024
• 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
3.5956	1.0	18600	3.7880	0.3589
3.3783	2.0	37200	3.5973	0.3816
3.2492	3.0	55800	3.4737	0.3920
3.174	4.0	74400	3.4102	0.3983
3.1174	5.0	93000	3.3910	0.4020
3.0763	6.0	111600	3.3897	0.4044
3.038	7.0	130200	3.3814	0.4058
3.0096	8.0	148800	3.3546	0.4082
2.9754	9.0	167400	3.3599	0.4077
2.9501	10.0	186000	3.3615	0.4083
2.9304	11.0	204600	3.3480	0.4101
2.9083	12.0	223200	3.3556	0.4097
2.8872	13.0	241800	3.3658	0.4093
2.8666	14.0	260400	3.3668	0.4103
2.841	15.0	279000	3.3685	0.4115
2.8247	16.0	297600	3.3752	0.4107
2.8096	17.0	316200	3.3834	0.4105
2.7845	18.0	334800	3.3846	0.4113
2.7699	19.0	353400	3.3916	0.4111
2.7549	20.0	372000	3.4015	0.4107

Framework versions

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