opt-babylm2-rewritten-clean-spacy_no-num-adj-earlystop-bpe_seed-42_1e-3

This model was trained from scratch on the kanishka/babylm2-rewritten-clean-spacy_no-num-adj dataset. It achieves the following results on the evaluation set:

• Loss: 2.6900
• Accuracy: 0.4781

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
4.0896	1.0	2225	3.8307	0.3593
3.4325	2.0	4450	3.3145	0.4081
3.1208	3.0	6675	3.1044	0.4295
2.957	4.0	8900	2.9973	0.4396
2.8381	5.0	11125	2.9338	0.4464
2.7819	6.0	13350	2.8904	0.4508
2.7385	7.0	15575	2.8666	0.4531
2.7061	8.0	17800	2.8456	0.4559
2.6855	9.0	20025	2.8332	0.4575
2.6669	10.0	22250	2.8198	0.4586
2.6499	11.0	24475	2.8118	0.4597
2.6351	12.0	26700	2.8072	0.4601
2.6204	13.0	28925	2.8026	0.4612
2.6277	14.0	31150	2.8013	0.4613
2.6136	15.0	33375	2.7791	0.4638
2.5687	16.0	35600	2.7514	0.4676
2.5184	17.0	37825	2.7283	0.4708
2.4613	18.0	40050	2.7060	0.4740
2.3966	19.0	42275	2.6947	0.4766
2.3227	19.9913	44480	2.6900	0.4781

Framework versions

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