smolm-autoreg-bpe-counterfactual-babylm-aann-prototypical_only-3e-4

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

• Loss: 3.4082
• Accuracy: 0.4080

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
3.7341	1.0	18593	3.8768	0.3468
3.4322	2.0	37186	3.6158	0.3751
3.2902	3.0	55779	3.4817	0.3883
3.21	4.0	74372	3.4286	0.3960
3.1498	5.0	92965	3.4151	0.3978
3.0981	6.0	111558	3.3790	0.4022
3.0651	7.0	130151	3.3750	0.4034
3.0292	8.0	148744	3.3879	0.4041
3.0031	9.0	167337	3.3773	0.4046
2.9713	10.0	185930	3.3769	0.4061
2.9529	11.0	204523	3.3778	0.4069
2.9286	12.0	223116	3.3612	0.4077
2.9065	13.0	241709	3.3686	0.4073
2.8837	14.0	260302	3.3861	0.4078
2.8621	15.0	278895	3.3851	0.4077
2.8487	16.0	297488	3.3876	0.4080
2.8243	17.0	316081	3.3908	0.4081
2.8078	18.0	334674	3.3952	0.4082
2.7887	19.0	353267	3.4020	0.4082
2.7716	20.0	371860	3.4082	0.4080

Framework versions

• Transformers 4.36.0
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0