nobinomial-babylm-base_seed-42_1e-4

This model was trained from scratch on the babylm_export dataset. It achieves the following results on the evaluation set:

• Loss: 3.1396
• Accuracy: 0.4068

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: 0.0001
• 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: 500
• num_epochs: 20.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.8857	1.0	1659	3.9011	0.3332
3.553	2.0	3318	3.5920	0.3604
3.4021	3.0	4977	3.4457	0.3744
3.2955	4.0	6636	3.3591	0.3832
3.2272	5.0	8295	3.3037	0.3888
3.1643	6.0	9954	3.2646	0.3926
3.1206	7.0	11613	3.2355	0.3957
3.0781	8.0	13272	3.2125	0.3980
3.0467	9.0	14931	3.1951	0.3999
3.023	10.0	16590	3.1822	0.4012
2.985	11.0	18249	3.1736	0.4023
2.9666	12.0	19908	3.1633	0.4035
2.9494	13.0	21567	3.1590	0.4041
2.9322	14.0	23226	3.1527	0.4048
2.9092	15.0	24885	3.1487	0.4054
2.8954	16.0	26544	3.1451	0.4059
2.8763	17.0	28203	3.1436	0.4061
2.8644	18.0	29862	3.1418	0.4065
2.8584	19.0	31521	3.1403	0.4067
2.8417	20.0	33180	3.1396	0.4068

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0