smolm-autoreg-bpe-counterfactual_babylm_aann_high_variability_numeral-seed_211-1e-3

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

• Loss: 3.4248
• Accuracy: 0.4100

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: 211
• 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
3.61	1.0	18596	3.7957	0.3584
3.3861	2.0	37192	3.5647	0.3801
3.2591	3.0	55788	3.4933	0.3915
3.1836	4.0	74384	3.4326	0.3972
3.1278	5.0	92980	3.4133	0.4018
3.0795	6.0	111576	3.4124	0.4033
3.0444	7.0	130172	3.3837	0.4052
3.0115	8.0	148768	3.3669	0.4068
2.9869	9.0	167364	3.3864	0.4073
2.9651	10.0	185960	3.3616	0.4079
2.9407	11.0	204556	3.3727	0.4090
2.9186	12.0	223152	3.3869	0.4091
2.8974	13.0	241748	3.3689	0.4096
2.8733	14.0	260344	3.3811	0.4098
2.8547	15.0	278940	3.3861	0.4102
2.8352	16.0	297536	3.3999	0.4099
2.8173	17.0	316132	3.3993	0.4104
2.8005	18.0	334728	3.4154	0.4099
2.7796	19.0	353324	3.4155	0.4102
2.7606	20.0	371920	3.4248	0.4100

Framework versions

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