smolm-autoreg-bpe-counterfactual-babylm-only_indef_articles_with_pl_nouns_removal-seed_1024-1e-3

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

• Loss: 3.4006
• Accuracy: 0.4110

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
3.6013	1.0	18600	3.7573	0.3598
3.3813	2.0	37200	3.5688	0.3805
3.2541	3.0	55800	3.4489	0.3922
3.174	4.0	74400	3.4158	0.3980
3.1166	5.0	93000	3.3767	0.4028
3.0777	6.0	111600	3.3729	0.4036
3.0372	7.0	130200	3.3464	0.4071
3.0083	8.0	148800	3.3503	0.4081
2.9762	9.0	167400	3.3317	0.4098
2.9515	10.0	186000	3.3434	0.4088
2.9338	11.0	204600	3.3526	0.4102
2.9063	12.0	223200	3.3577	0.4095
2.8871	13.0	241800	3.3493	0.4101
2.8654	14.0	260400	3.3641	0.4106
2.8465	15.0	279000	3.3597	0.4115
2.8233	16.0	297600	3.3748	0.4106
2.8071	17.0	316200	3.3754	0.4113
2.7899	18.0	334800	3.3833	0.4111
2.7669	19.0	353400	3.3913	0.4112
2.7513	20.0	372000	3.4006	0.4110

Framework versions

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