opt-babylm2-rewritten-clean-spacy-earlystop-bpe_seed-42_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: 2.6840
• 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: 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
4.1044	1.0	2256	3.8204	0.3604
3.4457	2.0	4512	3.3046	0.4093
3.13	3.0	6768	3.0945	0.4299
2.9219	4.0	9024	2.9890	0.4404
2.8444	5.0	11280	2.9282	0.4466
2.7883	6.0	13536	2.8910	0.4508
2.7434	7.0	15792	2.8579	0.4545
2.7158	8.0	18048	2.8428	0.4560
2.6905	9.0	20304	2.8298	0.4573
2.6697	10.0	22560	2.8169	0.4592
2.6509	11.0	24816	2.8080	0.4601
2.6494	12.0	27072	2.8020	0.4607
2.6384	13.0	29328	2.7958	0.4616
2.6297	14.0	31584	2.7939	0.4620
2.612	15.0	33840	2.7649	0.4653
2.5667	16.0	36096	2.7425	0.4686
2.5177	17.0	38352	2.7206	0.4714
2.4607	18.0	40608	2.6999	0.4746
2.397	19.0	42864	2.6865	0.4773
2.3241	19.9915	45100	2.6840	0.4787

Framework versions

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