4de24ea8b35aba44c5dc2d0b64ab3b91

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

• Loss: 0.8409
• Data Size: 1.0
• Epoch Runtime: 36.1938
• Mse: 0.8413
• Mae: 0.6921
• R2: 0.6237

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	7.4942	0	2.9076	7.4954	2.3046	-2.3530
No log	1	179	4.1056	0.0078	3.3955	4.1062	1.6384	-0.8368
No log	2	358	2.5640	0.0156	4.1054	2.5647	1.3208	-0.1473
No log	3	537	2.1231	0.0312	5.2480	2.1237	1.2032	0.0500
No log	4	716	1.3738	0.0625	6.7700	1.3745	0.9977	0.3852
No log	5	895	0.9138	0.125	9.2196	0.9144	0.7876	0.5910
0.1006	6	1074	1.3311	0.25	13.9642	1.3317	0.9130	0.4043
1.0011	7	1253	0.9213	0.5	22.1370	0.9217	0.7425	0.5877
0.9789	8.0	1432	0.7959	1.0	39.4309	0.7963	0.6983	0.6438
0.7642	9.0	1611	0.8431	1.0	37.5616	0.8435	0.7024	0.6227
0.7721	10.0	1790	0.8151	1.0	37.8112	0.8155	0.7126	0.6352
0.6567	11.0	1969	0.7596	1.0	38.4025	0.7599	0.6666	0.6601
0.6572	12.0	2148	0.8945	1.0	37.9143	0.8948	0.7292	0.5997
0.5637	13.0	2327	0.7626	1.0	37.4460	0.7630	0.6806	0.6587
0.5114	14.0	2506	0.8281	1.0	37.3292	0.8284	0.6919	0.6294
0.4406	15.0	2685	0.8409	1.0	36.1938	0.8413	0.6921	0.6237

Framework versions

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