smolm-autoreg-bpe-counterfactual-babylm-indef_articles_with_pl_nouns-removal-1e-3

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

• Loss: 3.4176
• Accuracy: 0.4125

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: 16
• 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
3.5148	1.0	37201	3.7270	0.3671
3.3074	2.0	74402	3.4841	0.3897
3.1988	3.0	111603	3.4300	0.3979
3.152	4.0	148804	3.3774	0.4050
3.0973	5.0	186005	3.3462	0.4090
3.0543	6.0	223206	3.3687	0.4064
3.0161	7.0	260407	3.3391	0.4114
2.9858	8.0	297608	3.3477	0.4105
2.9718	9.0	334809	3.3436	0.4112
2.9399	10.0	372010	3.3451	0.4121
2.9207	11.0	409211	3.3586	0.4130
2.8987	12.0	446412	3.3554	0.4123
2.8779	13.0	483613	3.3616	0.4130
2.8519	14.0	520814	3.3696	0.4129
2.8395	15.0	558015	3.3729	0.4128
2.8151	16.0	595216	3.3718	0.4140
2.798	17.0	632417	3.3858	0.4128
2.7738	18.0	669618	3.4080	0.4130
2.7555	19.0	706819	3.4067	0.4131
2.7434	20.0	744020	3.4176	0.4125

Framework versions

• Transformers 4.36.0
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0