d0946ad6815c8c1dd331577690f1d309

This model is a fine-tuned version of studio-ousia/luke-japanese-base on the nyu-mll/glue [stsb] dataset. It achieves the following results on the evaluation set:

• Loss: 0.8136
• Data Size: 1.0
• Epoch Runtime: 18.5317
• Mse: 0.8139
• Mae: 0.6826
• R2: 0.6359

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	Mse	Mae	R2
No log	0	0	5.8981	0	1.8152	5.8993	2.0135	-1.6390
No log	1	179	2.3333	0.0078	2.2352	2.3342	1.3199	-0.0442
No log	2	358	2.5393	0.0156	2.4516	2.5400	1.3217	-0.1362
No log	3	537	2.1605	0.0312	2.8056	2.1612	1.2158	0.0332
No log	4	716	1.7687	0.0625	3.7649	1.7691	1.0172	0.2086
No log	5	895	1.1003	0.125	4.8004	1.1007	0.8125	0.5076
0.1041	6	1074	1.0641	0.25	7.8165	1.0647	0.7867	0.5237
1.0771	7	1253	0.9305	0.5	10.8008	0.9309	0.7687	0.5836
0.8551	8.0	1432	0.7408	1.0	18.8081	0.7411	0.6798	0.6685
0.6673	9.0	1611	0.8166	1.0	18.5559	0.8170	0.6850	0.6345
0.6015	10.0	1790	0.8026	1.0	19.4479	0.8029	0.7005	0.6408
0.4784	11.0	1969	0.9393	1.0	18.2009	0.9397	0.7365	0.5796
0.4984	12.0	2148	0.8136	1.0	18.5317	0.8139	0.6826	0.6359

Framework versions

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