opt-babylm2-rewritten-clean-spacy-earlystop_hierarchical_211_age-origin_adj1-bpe_seed-1024_1e-3

This model was trained from scratch on the kanishka/babylm2-rewritten-clean-spacy_hierarchical-adj_211_age-origin_adj1-ablation dataset. It achieves the following results on the evaluation set:

• Loss: 2.6835
• Accuracy: 0.4786

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
4.0968	1.0	2256	3.8140	0.3606
3.452	2.0	4512	3.3035	0.4093
3.1308	3.0	6768	3.0885	0.4307
2.9235	4.0	9024	2.9829	0.4410
2.8434	5.0	11280	2.9234	0.4472
2.7855	6.0	13536	2.8829	0.4515
2.7432	7.0	15792	2.8552	0.4545
2.7116	8.0	18048	2.8346	0.4570
2.6898	9.0	20304	2.8246	0.4582
2.6685	10.0	22560	2.8111	0.4599
2.6486	11.0	24816	2.8039	0.4605
2.6485	12.0	27072	2.7972	0.4611
2.6393	13.0	29328	2.7920	0.4619
2.6275	14.0	31584	2.7878	0.4621
2.611	15.0	33840	2.7655	0.4651
2.5695	16.0	36096	2.7403	0.4686
2.5194	17.0	38352	2.7194	0.4715
2.4643	18.0	40608	2.6999	0.4746
2.398	19.0	42864	2.6852	0.4772
2.3252	20.0	45120	2.6835	0.4786

Framework versions

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