smolm-autoreg-bpe-counterfactual-babylm-only_measure_nps_as_singular_removal-seed_211-1e-3

This model was trained from scratch on the kanishka/counterfactual-babylm-only_measure_nps_as_singular_removal dataset. It achieves the following results on the evaluation set:

• Loss: 3.4372
• Accuracy: 0.4092

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
3.6018	1.0	18600	3.7779	0.3590
3.3799	2.0	37200	3.5990	0.3799
3.2535	3.0	55800	3.4629	0.3928
3.1731	4.0	74400	3.4447	0.3979
3.1186	5.0	93000	3.4295	0.4009
3.0776	6.0	111600	3.4004	0.4034
3.0407	7.0	130200	3.3850	0.4053
3.0066	8.0	148800	3.3648	0.4061
2.9851	9.0	167400	3.3985	0.4074
2.953	10.0	186000	3.3964	0.4077
2.9321	11.0	204600	3.3816	0.4088
2.9082	12.0	223200	3.3780	0.4093
2.8881	13.0	241800	3.4020	0.4090
2.8698	14.0	260400	3.4057	0.4091
2.8441	15.0	279000	3.3906	0.4094
2.8256	16.0	297600	3.4051	0.4094
2.808	17.0	316200	3.4108	0.4093
2.7945	18.0	334800	3.4283	0.4094
2.7744	19.0	353400	3.4362	0.4094
2.7567	20.0	372000	3.4372	0.4092

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1