opt-babylm2-rewritten-clean-spacy-earlystop_hierarchica_size-origin_both-bpe_seed-42_1e-3

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

• Loss: 2.6840
• Accuracy: 0.4785

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
4.0972	1.0	2256	3.8128	0.3609
3.4452	2.0	4512	3.2994	0.4094
3.1255	3.0	6768	3.0912	0.4298
2.9212	4.0	9024	2.9858	0.4406
2.8437	5.0	11280	2.9291	0.4465
2.788	6.0	13536	2.8833	0.4512
2.7431	7.0	15792	2.8611	0.4538
2.7134	8.0	18048	2.8409	0.4563
2.6877	9.0	20304	2.8257	0.4579
2.6686	10.0	22560	2.8134	0.4594
2.6516	11.0	24816	2.8029	0.4607
2.6489	12.0	27072	2.7985	0.4614
2.6394	13.0	29328	2.7963	0.4615
2.6281	14.0	31584	2.7905	0.4621
2.612	15.0	33840	2.7675	0.4650
2.5673	16.0	36096	2.7418	0.4682
2.5202	17.0	38352	2.7217	0.4713
2.4626	18.0	40608	2.7019	0.4745
2.3999	19.0	42864	2.6872	0.4768
2.3249	20.0	45120	2.6840	0.4785

Framework versions

• Transformers 4.48.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.4