opt-babylm2-rewritten-clean-spacy_no-num-adj-earlystop-bpe_seed-1024_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: 2.6899
• 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: 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
4.0745	1.0	2225	3.8237	0.3601
3.4383	2.0	4450	3.3184	0.4082
3.1188	3.0	6675	3.1054	0.4287
2.9576	4.0	8900	2.9992	0.4395
2.8402	5.0	11125	2.9343	0.4458
2.7813	6.0	13350	2.8921	0.4506
2.7354	7.0	15575	2.8686	0.4534
2.7041	8.0	17800	2.8453	0.4558
2.6834	9.0	20025	2.8331	0.4573
2.6649	10.0	22250	2.8191	0.4589
2.6483	11.0	24475	2.8133	0.4598
2.632	12.0	26700	2.8044	0.4605
2.6182	13.0	28925	2.7993	0.4612
2.6262	14.0	31150	2.7950	0.4619
2.6128	15.0	33375	2.7751	0.4643
2.5679	16.0	35600	2.7503	0.4677
2.5149	17.0	37825	2.7277	0.4710
2.4598	18.0	40050	2.7076	0.4741
2.3955	19.0	42275	2.6928	0.4767
2.3212	19.9913	44480	2.6899	0.4781

Framework versions

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