smolm-autoreg-bpe-counterfactual_babylm_aann_indef_articles_with_pl_nouns_removal_new-3e-4

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: 3.4250
• Accuracy: 0.4092

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
3.7386	1.0	18600	3.9390	0.3446
3.4323	2.0	37200	3.6439	0.3747
3.2906	3.0	55800	3.5132	0.3878
3.2008	4.0	74400	3.4662	0.3952
3.1424	5.0	93000	3.4252	0.3988
3.0983	6.0	111600	3.4146	0.4023
3.061	7.0	130200	3.3961	0.4039
3.0241	8.0	148800	3.3675	0.4061
2.9955	9.0	167400	3.3690	0.4071
2.971	10.0	186000	3.3668	0.4077
2.9425	11.0	204600	3.3717	0.4083
2.9175	12.0	223200	3.3836	0.4085
2.8993	13.0	241800	3.3685	0.4096
2.8802	14.0	260400	3.3869	0.4094
2.8591	15.0	279000	3.3903	0.4093
2.8397	16.0	297600	3.3899	0.4099
2.8158	17.0	316200	3.3992	0.4095
2.7994	18.0	334800	3.4129	0.4090
2.7773	19.0	353400	3.4211	0.4092
2.7599	20.0	372000	3.4250	0.4092

Framework versions

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