smolm-autoreg-bpe-counterfactual_babylm_aann_high_variability_adj-seed_1024-1e-3

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

• Loss: 3.4042
• Accuracy: 0.4105

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: 1024
• 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
3.5971	1.0	18594	3.7361	0.3618
3.3831	2.0	37188	3.5640	0.3810
3.2562	3.0	55782	3.4518	0.3928
3.1823	4.0	74376	3.4151	0.3979
3.123	5.0	92970	3.3895	0.4020
3.0779	6.0	111564	3.3728	0.4045
3.0394	7.0	130158	3.3513	0.4068
3.0098	8.0	148752	3.3329	0.4091
2.9848	9.0	167346	3.3543	0.4096
2.9607	10.0	185940	3.3317	0.4102
2.9381	11.0	204534	3.3509	0.4096
2.9132	12.0	223128	3.3383	0.4106
2.8884	13.0	241722	3.3772	0.4105
2.8698	14.0	260316	3.3457	0.4117
2.8512	15.0	278910	3.3592	0.4110
2.828	16.0	297504	3.3782	0.4106
2.813	17.0	316098	3.3778	0.4109
2.7978	18.0	334692	3.3931	0.4105
2.7756	19.0	353286	3.3947	0.4107
2.7571	20.0	371880	3.4042	0.4105

Framework versions

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