98ea2e6bed01827f3a1ca3ff6313497c

This model is a fine-tuned version of studio-ousia/luke-japanese-base on the contemmcm/trec dataset. It achieves the following results on the evaluation set:

• Loss: 0.6920
• Data Size: 1.0
• Epoch Runtime: 17.4328
• Accuracy: 0.8729
• F1 Macro: 0.8785

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	1.8820	0	0.9298	0.0187	0.0061
No log	1	170	1.6869	0.0078	1.4200	0.2771	0.0723
No log	2	340	1.7056	0.0156	1.5828	0.2708	0.1529
No log	3	510	1.5791	0.0312	1.9501	0.3417	0.1788
No log	4	680	1.3250	0.0625	2.6109	0.5021	0.3641
0.0874	5	850	1.1021	0.125	3.9563	0.6062	0.5003
0.0874	6	1020	0.7524	0.25	5.8304	0.7375	0.6224
0.8323	7	1190	0.6923	0.5	9.8112	0.7729	0.6255
0.5854	8.0	1360	0.5847	1.0	18.0356	0.8083	0.8035
0.4273	9.0	1530	0.3785	1.0	17.3681	0.8792	0.8733
0.3629	10.0	1700	0.5134	1.0	18.5573	0.8479	0.8539
0.2921	11.0	1870	0.3418	1.0	17.5902	0.8958	0.8992
0.197	12.0	2040	0.4025	1.0	17.3387	0.8792	0.8866
0.1747	13.0	2210	0.5681	1.0	18.3054	0.8708	0.8837
0.1765	14.0	2380	0.4541	1.0	17.4103	0.8896	0.8954
0.1162	15.0	2550	0.6920	1.0	17.4328	0.8729	0.8785

Framework versions

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