opt-babylm2-rewritten-clean-spacy-32k-earlystop-40epochs_seed-42_1e-3

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

• Loss: 2.9600
• Accuracy: 0.4233

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 32000
• num_epochs: 40.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
5.9216	0.9996	1931	4.0134	0.3253
3.7977	1.9997	3863	3.5448	0.3639
3.3887	2.9999	5795	3.3242	0.3841
3.1805	4.0	7727	3.2082	0.3949
3.0632	4.9996	9658	3.1432	0.4012
2.9865	5.9997	11590	3.1010	0.4056
2.9347	6.9999	13522	3.0715	0.4087
2.8953	8.0	15454	3.0539	0.4108
2.8689	8.9996	17385	3.0392	0.4122
2.8456	9.9997	19317	3.0310	0.4134
2.8298	10.9999	21249	3.0251	0.4144
2.817	12.0	23181	3.0175	0.4152
2.8069	12.9996	25112	3.0119	0.4158
2.7996	13.9997	27044	3.0060	0.4163
2.7615	14.9999	28976	3.0038	0.4171
2.7575	16.0	30908	3.0022	0.4169
2.7573	16.9996	32839	2.9962	0.4179
2.7451	17.9997	34771	2.9867	0.4189
2.7275	18.9999	36703	2.9804	0.4201
2.7099	20.0	38635	2.9760	0.4208
2.693	20.9996	40566	2.9683	0.4216
2.6785	21.9997	42498	2.9666	0.4221
2.6628	22.9999	44430	2.9646	0.4227
2.6501	24.0	46362	2.9626	0.4228
2.6343	24.9996	48293	2.9600	0.4233
2.6198	25.9997	50225	2.9638	0.4236
2.604	26.9999	52157	2.9604	0.4240
2.5876	28.0	54089	2.9601	0.4245

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0