dc503a298dc7d8be342da43ad388cd77

This model is a fine-tuned version of google-bert/bert-base-german-cased on the contemmcm/cls_mmlu dataset. It achieves the following results on the evaluation set:

• Loss: 1.3884
• Data Size: 1.0
• Epoch Runtime: 23.8602
• Accuracy: 0.2527
• F1 Macro: 0.1008

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Data Size	Epoch Runtime	Accuracy	F1 Macro
No log	0	0	1.4089	0	1.2732	0.2487	0.0996
No log	1	438	1.4563	0.0078	1.7745	0.2527	0.1008
No log	2	876	1.4450	0.0156	1.9226	0.2453	0.0985
No log	3	1314	1.4318	0.0312	2.3163	0.2540	0.1024
No log	4	1752	1.4016	0.0625	3.1471	0.2507	0.1139
0.0787	5	2190	1.4057	0.125	4.6377	0.2527	0.1059
0.1852	6	2628	1.4000	0.25	7.3769	0.2507	0.1038
1.4006	7	3066	1.3897	0.5	12.9338	0.2533	0.1347
1.395	8.0	3504	1.3927	1.0	24.6712	0.2527	0.1008
1.3954	9.0	3942	1.4000	1.0	23.6264	0.2533	0.1011
1.3949	10.0	4380	1.3881	1.0	23.6541	0.2527	0.1008
1.3923	11.0	4818	1.3897	1.0	23.9062	0.2527	0.1008
1.3943	12.0	5256	1.3898	1.0	23.9248	0.2527	0.1008
1.3888	13.0	5694	1.3906	1.0	23.7377	0.2533	0.1011
1.3951	14.0	6132	1.3884	1.0	23.8602	0.2527	0.1008

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu128
• Datasets 4.3.0
• Tokenizers 0.22.1