308cf0de4623ecf2177f369cfef38cde

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

• Loss: 0.3553
• Data Size: 1.0
• Epoch Runtime: 290.0445
• Accuracy: 0.8824
• F1 Macro: 0.8823
• Rouge1: 0.8822
• Rouge2: 0.0
• Rougel: 0.8824
• Rougelsum: 0.8825

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	0.6937	0	5.1626	0.4941	0.3314	0.4946	0.0	0.4941	0.4943
No log	1	3273	0.5359	0.0078	7.7174	0.7649	0.7613	0.7651	0.0	0.7647	0.7647
0.0099	2	6546	0.4260	0.0156	9.8649	0.8311	0.8299	0.8309	0.0	0.8313	0.8313
0.4672	3	9819	0.4062	0.0312	14.5216	0.8320	0.8311	0.8320	0.0	0.8320	0.8320
0.4012	4	13092	0.3313	0.0625	24.4607	0.8588	0.8585	0.8588	0.0	0.8588	0.8588
0.3272	5	16365	0.3435	0.125	40.5947	0.8654	0.8652	0.8653	0.0	0.8654	0.8651
0.3353	6	19638	0.2847	0.25	75.4992	0.8816	0.8815	0.8816	0.0	0.8818	0.8818
0.3175	7	22911	0.2803	0.5	147.2464	0.8989	0.8989	0.8987	0.0	0.8989	0.8989
0.3174	8.0	26184	0.3178	1.0	287.3263	0.8943	0.8942	0.8943	0.0	0.8943	0.8941
0.3027	9.0	29457	0.3085	1.0	285.6588	0.8851	0.8850	0.8849	0.0	0.8850	0.8850
0.2493	10.0	32730	0.3197	1.0	285.1441	0.8949	0.8948	0.8947	0.0	0.8949	0.8949
0.2529	11.0	36003	0.3553	1.0	290.0445	0.8824	0.8823	0.8822	0.0	0.8824	0.8825

Framework versions

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