smolm-autoreg-bpe-counterfactual_babylm_aann_high_variability_numeral-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: 3.3843
• Accuracy: 0.4119

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
3.5998	1.0	18596	3.7535	0.3595
3.383	2.0	37192	3.5660	0.3813
3.258	3.0	55788	3.5104	0.3914
3.1754	4.0	74384	3.4088	0.3985
3.1222	5.0	92980	3.4039	0.4021
3.0807	6.0	111576	3.3850	0.4050
3.0411	7.0	130172	3.3609	0.4064
3.0096	8.0	148768	3.3502	0.4082
2.9845	9.0	167364	3.3612	0.4089
2.9585	10.0	185960	3.3495	0.4091
2.9308	11.0	204556	3.3414	0.4115
2.9139	12.0	223152	3.3366	0.4117
2.8894	13.0	241748	3.3379	0.4112
2.8685	14.0	260344	3.3566	0.4113
2.8472	15.0	278940	3.3577	0.4114
2.827	16.0	297536	3.3584	0.4123
2.8091	17.0	316132	3.3707	0.4115
2.7903	18.0	334728	3.3751	0.4117
2.7681	19.0	353324	3.3758	0.4121
2.7544	20.0	371920	3.3843	0.4119

Framework versions

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