smolm-autoreg-bpe-counterfactual_babylm_naans_new-seed_211-1e-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: 3.4159
• Accuracy: 0.4060

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: 0.0001
• 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
4.0574	1.0	18595	4.2668	0.3095
3.574	2.0	37190	3.7340	0.3638
3.3998	3.0	55785	3.5946	0.3793
3.2908	4.0	74380	3.5169	0.3871
3.2244	5.0	92975	3.4843	0.3919
3.1723	6.0	111570	3.4423	0.3955
3.1287	7.0	130165	3.4224	0.3987
3.0995	8.0	148760	3.4119	0.4001
3.0666	9.0	167355	3.4093	0.4014
3.0395	10.0	185950	3.3993	0.4024
3.0097	11.0	204545	3.4087	0.4031
2.9923	12.0	223140	3.4030	0.4042
2.9703	13.0	241735	3.3938	0.4047
2.9483	14.0	260330	3.4000	0.4051
2.9286	15.0	278925	3.4069	0.4048
2.9143	16.0	297520	3.4020	0.4056
2.8935	17.0	316115	3.4100	0.4055
2.8782	18.0	334710	3.4071	0.4058
2.8613	19.0	353305	3.4123	0.4062
2.8439	20.0	371900	3.4159	0.4060

Framework versions

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