smolm-autoreg-bpe-counterfactual_babylm_naans_new-3e-4

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

• Loss: 3.4288
• 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
3.7356	1.0	18595	3.8848	0.3475
3.4297	2.0	37190	3.6216	0.3763
3.2928	3.0	55785	3.5005	0.3889
3.2003	4.0	74380	3.4656	0.3952
3.1447	5.0	92975	3.4068	0.3994
3.0991	6.0	111570	3.4298	0.4022
3.0592	7.0	130165	3.3990	0.4041
3.0279	8.0	148760	3.3796	0.4057
2.9978	9.0	167355	3.3757	0.4063
2.9761	10.0	185950	3.3907	0.4068
2.9464	11.0	204545	3.3881	0.4073
2.9236	12.0	223140	3.3905	0.4080
2.907	13.0	241735	3.3880	0.4087
2.8803	14.0	260330	3.3927	0.4088
2.8614	15.0	278925	3.3924	0.4091
2.8414	16.0	297520	3.3970	0.4098
2.8226	17.0	316115	3.4111	0.4092
2.8063	18.0	334710	3.4199	0.4091
2.7905	19.0	353305	3.4234	0.4093
2.7753	20.0	371900	3.4288	0.4092

Framework versions

• Transformers 4.40.1
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.19.1