smolm-autoreg-bpe-counterfactual_babylm_aann_low_variability_numeral-1e-3

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

• Loss: 3.4341
• 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
3.6001	1.0	18593	3.7711	0.3593
3.383	2.0	37186	3.6063	0.3805
3.2545	3.0	55779	3.4785	0.3921
3.1841	4.0	74372	3.4519	0.3980
3.128	5.0	92965	3.4033	0.4010
3.0818	6.0	111558	3.4102	0.4041
3.0463	7.0	130151	3.3928	0.4052
3.0171	8.0	148744	3.3537	0.4073
2.9883	9.0	167337	3.3732	0.4075
2.9627	10.0	185930	3.3831	0.4084
2.9352	11.0	204523	3.3770	0.4089
2.9164	12.0	223116	3.3897	0.4088
2.8914	13.0	241709	3.3744	0.4094
2.8738	14.0	260302	3.3724	0.4098
2.8534	15.0	278895	3.3984	0.4100
2.8347	16.0	297488	3.4039	0.4097
2.8177	17.0	316081	3.4102	0.4099
2.7952	18.0	334674	3.4160	0.4098
2.7803	19.0	353267	3.4237	0.4099
2.7629	20.0	371860	3.4341	0.4098

Framework versions

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