smolm-autoreg-bpe-counterfactual_babylm_aann_low_variability_noun_211-1e-3

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

  • Loss: 3.4073
  • Accuracy: 0.4102

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
3.6045 1.0 18595 3.8020 0.3570
3.3841 2.0 37190 3.5977 0.3801
3.2569 3.0 55785 3.4644 0.3932
3.1775 4.0 74380 3.4314 0.3981
3.1198 5.0 92975 3.4138 0.4017
3.0787 6.0 111570 3.3922 0.4036
3.0411 7.0 130165 3.3671 0.4059
3.0125 8.0 148760 3.3583 0.4072
2.9812 9.0 167355 3.3392 0.4088
2.9592 10.0 185950 3.3605 0.4087
2.9333 11.0 204545 3.3525 0.4092
2.9129 12.0 223140 3.3609 0.4097
2.8907 13.0 241735 3.3751 0.4097
2.8737 14.0 260330 3.3661 0.4099
2.8533 15.0 278925 3.3723 0.4102
2.8283 16.0 297520 3.3770 0.4106
2.817 17.0 316115 3.3919 0.4102
2.7925 18.0 334710 3.3901 0.4106
2.7778 19.0 353305 3.4038 0.4102
2.7587 20.0 371900 3.4073 0.4102

Framework versions

  • Transformers 4.38.0
  • Pytorch 2.3.1+cu121
  • Datasets 2.16.1
  • Tokenizers 0.15.2
Downloads last month
8
Safetensors
Model size
97.8M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train kanishka/smolm-autoreg-bpe-counterfactual_babylm_aann_low_variability_noun-seed_211-1e-3

Evaluation results

  • Accuracy on kanishka/counterfactual_babylm_aann_low_variability_noun
    self-reported
    0.410