opt-babylm2-clean-spacy-earlystop-bpe_seed-1024_1e-3

This model was trained from scratch on the kanishka/babylm2-clean-spacy dataset. It achieves the following results on the evaluation set:

• Loss: 2.6779
• Accuracy: 0.4789

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
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• 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
4.0976	1.0	2264	3.8103	0.3607
3.4447	2.0	4528	3.2977	0.4097
3.1266	3.0	6792	3.0873	0.4300
2.9206	4.0	9056	2.9802	0.4412
2.839	5.0	11320	2.9201	0.4471
2.785	6.0	13584	2.8801	0.4516
2.7376	7.0	15848	2.8532	0.4545
2.7088	8.0	18112	2.8310	0.4569
2.6833	9.0	20376	2.8196	0.4584
2.6631	10.0	22640	2.8089	0.4595
2.6439	11.0	24904	2.8001	0.4604
2.6426	12.0	27168	2.7931	0.4613
2.632	13.0	29432	2.7888	0.4621
2.6204	14.0	31696	2.7885	0.4617
2.6043	15.0	33960	2.7619	0.4654
2.5616	16.0	36224	2.7378	0.4687
2.5124	17.0	38488	2.7151	0.4718
2.4554	18.0	40752	2.6965	0.4749
2.3891	19.0	43016	2.6810	0.4776
2.321	19.9914	45260	2.6779	0.4789

Framework versions

• Transformers 4.48.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.1