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

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

• Loss: 2.6845
• Accuracy: 0.4787

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
4.093	1.0	2256	3.8133	0.3607
3.4493	2.0	4512	3.3059	0.4091
3.132	3.0	6768	3.0989	0.4295
2.9232	4.0	9024	2.9917	0.4401
2.8445	5.0	11280	2.9270	0.4466
2.7872	6.0	13536	2.8861	0.4510
2.7438	7.0	15792	2.8637	0.4541
2.7139	8.0	18048	2.8414	0.4565
2.6887	9.0	20304	2.8274	0.4580
2.6686	10.0	22560	2.8194	0.4590
2.65	11.0	24816	2.8066	0.4605
2.6481	12.0	27072	2.8002	0.4611
2.6382	13.0	29328	2.7956	0.4618
2.6285	14.0	31584	2.7968	0.4619
2.6116	15.0	33840	2.7692	0.4650
2.5677	16.0	36096	2.7446	0.4684
2.5197	17.0	38352	2.7191	0.4717
2.462	18.0	40608	2.7010	0.4747
2.3973	19.0	42864	2.6860	0.4773
2.324	19.9915	45100	2.6845	0.4787

Framework versions

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