opt-babylm2-rewritten-clean-spacy_no-num-adj-earlystop-bpe_seed-42_1e-3

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

• Loss: 2.6889
• Accuracy: 0.4781

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: 42
• 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
4.0804	1.0	2226	3.8291	0.3599
3.431	2.0	4452	3.3167	0.4076
3.1165	3.0	6678	3.1028	0.4290
2.9545	4.0	8904	2.9982	0.4396
2.8357	5.0	11130	2.9296	0.4463
2.7805	6.0	13356	2.8928	0.4505
2.7354	7.0	15582	2.8630	0.4537
2.702	8.0	17808	2.8441	0.4561
2.684	9.0	20034	2.8353	0.4571
2.6647	10.0	22260	2.8199	0.4584
2.6475	11.0	24486	2.8120	0.4596
2.6325	12.0	26712	2.8057	0.4604
2.6204	13.0	28938	2.7991	0.4611
2.6273	14.0	31164	2.7976	0.4613
2.6134	15.0	33390	2.7782	0.4638
2.5675	16.0	35616	2.7498	0.4674
2.5158	17.0	37842	2.7294	0.4708
2.4606	18.0	40068	2.7055	0.4738
2.3953	19.0	42294	2.6920	0.4765
2.3225	19.9913	44500	2.6889	0.4781

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1
• Datasets 3.2.0
• Tokenizers 0.21.0