opt-babylm2-rewritten-clean-spacy_no-num-adj-earlystop-bpe_seed-1024_1e-3

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

• Loss: 2.6427
• Accuracy: 0.4823

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
• 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
3.1046	1.0	17787	3.1574	0.4223
2.9659	2.0	35574	3.0541	0.4326
2.9018	3.0	53361	2.9921	0.4386
2.858	4.0	71148	2.9562	0.4425
2.8303	5.0	88935	2.9372	0.4447
2.8174	6.0	106722	2.9199	0.4469
2.7932	7.0	124509	2.9019	0.4488
2.776	8.0	142296	2.8863	0.4503
2.7565	9.0	160083	2.8733	0.4519
2.7316	10.0	177870	2.8549	0.4541
2.7164	11.0	195657	2.8356	0.4565
2.7009	12.0	213444	2.8189	0.4583
2.6774	13.0	231231	2.8008	0.4607
2.6515	14.0	249018	2.7832	0.4627
2.6173	15.0	266805	2.7613	0.4654
2.5889	16.0	284592	2.7395	0.4682
2.5406	17.0	302379	2.7126	0.4714
2.4929	18.0	320166	2.6857	0.4753
2.4224	19.0	337953	2.6585	0.4793
2.3296	20.0	355740	2.6427	0.4823

Framework versions

• Transformers 4.48.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.1