smolm-autoreg-bpe-counterfactual_babylm_anans_new-3e-4

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

• Loss: 3.4257
• Accuracy: 0.4096

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
3.7355	1.0	18595	3.8995	0.3461
3.4292	2.0	37190	3.6087	0.3768
3.2927	3.0	55785	3.5001	0.3895
3.1999	4.0	74380	3.4743	0.3952
3.1448	5.0	92975	3.4089	0.3996
3.0986	6.0	111570	3.4166	0.4022
3.0592	7.0	130165	3.4032	0.4036
3.0279	8.0	148760	3.3746	0.4062
2.9977	9.0	167355	3.3709	0.4068
2.9761	10.0	185950	3.3795	0.4071
2.9464	11.0	204545	3.3783	0.4080
2.9234	12.0	223140	3.3832	0.4084
2.9068	13.0	241735	3.3838	0.4087
2.88	14.0	260330	3.3881	0.4091
2.8614	15.0	278925	3.3863	0.4097
2.841	16.0	297520	3.4092	0.4094
2.8225	17.0	316115	3.3966	0.4098
2.8062	18.0	334710	3.4095	0.4096
2.7904	19.0	353305	3.4169	0.4098
2.775	20.0	371900	3.4257	0.4096

Framework versions

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