opt-babylm2-rewritten-clean-spacy-earlystop-long-bpe_seed-211_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: 2.6405
• Accuracy: 0.4824

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: 211
• 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
3.0875	1.0	18042	3.1520	0.4227
2.9734	2.0	36084	3.0482	0.4331
2.9048	3.0	54126	2.9848	0.4394
2.8742	4.0	72168	2.9526	0.4430
2.8494	5.0	90210	2.9296	0.4458
2.819	6.0	108252	2.9178	0.4471
2.8034	7.0	126294	2.8993	0.4489
2.7878	8.0	144336	2.8818	0.4511
2.7688	9.0	162378	2.8669	0.4529
2.7492	10.0	180420	2.8508	0.4546
2.7289	11.0	198462	2.8333	0.4566
2.7045	12.0	216504	2.8186	0.4583
2.682	13.0	234546	2.8000	0.4605
2.6581	14.0	252588	2.7817	0.4626
2.6307	15.0	270630	2.7613	0.4649
2.5966	16.0	288672	2.7396	0.4682
2.5502	17.0	306714	2.7122	0.4716
2.4952	18.0	324756	2.6853	0.4752
2.4283	19.0	342798	2.6560	0.4795
2.3386	20.0	360840	2.6405	0.4824

Framework versions

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