Edit model card

smolm-autoreg-bpe-counterfactual-babylm-new_regex_aanns_removal-1e-4

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

  • Loss: 3.3971
  • Accuracy: 0.4081

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: 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
4.0555 1.0 18593 4.2608 0.3086
3.5626 2.0 37186 3.7490 0.3634
3.3926 3.0 55779 3.5729 0.3804
3.2863 4.0 74372 3.5122 0.3883
3.2223 5.0 92965 3.4704 0.3932
3.1687 6.0 111558 3.4542 0.3960
3.1265 7.0 130151 3.4307 0.3987
3.0955 8.0 148744 3.4010 0.4014
3.0614 9.0 167337 3.3947 0.4026
3.0346 10.0 185930 3.3861 0.4037
3.0121 11.0 204523 3.3788 0.4047
2.9917 12.0 223116 3.3737 0.4050
2.968 13.0 241709 3.3828 0.4055
2.9462 14.0 260302 3.3921 0.4060
2.9308 15.0 278895 3.3842 0.4073
2.9096 16.0 297488 3.3800 0.4075
2.889 17.0 316081 3.3850 0.4077
2.8779 18.0 334674 3.3920 0.4076
2.8585 19.0 353267 3.3898 0.4084
2.8469 20.0 371860 3.3971 0.4081

Framework versions

  • Transformers 4.40.1
  • Pytorch 2.1.0+cu121
  • Datasets 2.16.1
  • Tokenizers 0.19.1
Downloads last month
1
Safetensors
Model size
97.8M params
Tensor type
F32
·
Inference API
This model can be loaded on Inference API (serverless).

Dataset used to train kanishka/smolm-autoreg-bpe-counterfactual-babylm-new_regex_aanns_removal-1e-4

Evaluation results

  • Accuracy on kanishka/counterfactual-babylm-new_regex_aanns_removal
    self-reported
    0.408