babylm-base10m-gpt2

This model is a fine-tuned version of pretrain_service/decoder/config/config.json on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.8694
• Accuracy: 0.4824

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: 5e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• 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: 200
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
5.1122	0.0909	200	4.7293	0.3483
4.3375	0.1817	400	4.2771	0.3642
4.1386	0.2726	600	4.1027	0.3726
4.0068	0.3635	800	4.0034	0.3750
3.8941	0.4543	1000	3.9384	0.3786
3.778	0.5452	1200	3.8843	0.3818
3.7554	0.6361	1400	3.8389	0.3851
3.6796	0.7269	1600	3.7815	0.3915
3.6266	0.8178	1800	3.7347	0.3952
3.601	0.9087	2000	3.6784	0.4023
3.2525	1.8174	4000	3.3798	0.4285
3.1444	2.7260	6000	3.2090	0.4433
2.9805	3.6347	8000	3.0989	0.4543
2.8567	4.5434	10000	3.0173	0.4636
2.7515	5.4521	12000	2.9567	0.4713
2.7002	6.3607	14000	2.9195	0.4757
2.6189	7.2694	16000	2.8929	0.4792
2.6109	8.1781	18000	2.8772	0.4813

Framework versions

• Transformers 4.50.3
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.4