opt-babylm2-rewritten-clean-spacy-earlystop_hierarchica_size-origin_adj1-bpe_seed-42_1e-3

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

• Loss: 2.6838
• 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
4.0948	1.0	2257	3.8187	0.3602
3.4457	2.0	4514	3.3142	0.4069
3.1274	3.0	6771	3.1016	0.4289
2.9251	4.0	9028	2.9907	0.4401
2.8428	5.0	11285	2.9315	0.4460
2.7861	6.0	13542	2.8903	0.4507
2.7424	7.0	15799	2.8599	0.4536
2.7122	8.0	18056	2.8416	0.4559
2.6878	9.0	20313	2.8270	0.4578
2.6692	10.0	22570	2.8142	0.4589
2.6513	11.0	24827	2.8101	0.4599
2.6494	12.0	27084	2.8005	0.4601
2.6369	13.0	29341	2.8019	0.4608
2.6283	14.0	31598	2.7923	0.4617
2.6113	15.0	33855	2.7735	0.4645
2.5694	16.0	36112	2.7425	0.4682
2.5184	17.0	38369	2.7224	0.4711
2.4614	18.0	40626	2.7012	0.4743
2.3979	19.0	42883	2.6872	0.4772
2.3255	19.9915	45120	2.6838	0.4785

Framework versions

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