fc9bf5cb1e70b83309cc1d874d86a17e

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

• Loss: 1.0974
• Data Size: 0.125
• Epoch Runtime: 137.9579
• Accuracy: 0.3545
• F1 Macro: 0.1745
• Rouge1: 0.3544
• Rouge2: 0.0
• Rougel: 0.3545
• Rougelsum: 0.3543

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	Rouge1	Rouge2	Rougel	Rougelsum
No log	0	0	1.1264	0	8.6970	0.3545	0.1745	0.3544	0.0	0.3545	0.3543
1.1148	1	12271	1.0498	0.0078	17.1529	0.4245	0.3902	0.4247	0.0	0.4247	0.4246
1.1052	2	24542	1.0974	0.0156	25.3839	0.3545	0.1745	0.3544	0.0	0.3545	0.3543
1.1042	3	36813	1.0975	0.0312	41.3318	0.3273	0.1644	0.3273	0.0	0.3275	0.3277
1.1045	4	49084	1.0989	0.0625	74.8124	0.3545	0.1745	0.3544	0.0	0.3545	0.3543
1.103	5	61355	1.0974	0.125	137.9579	0.3545	0.1745	0.3544	0.0	0.3545	0.3543

Framework versions

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