squad-bn-mt5-base2

This model is a fine-tuned version of google/mt5-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5309
• Rouge1 Precision: 37.5039
• Rouge1 Recall: 30.4476
• Rouge1 Fmeasure: 32.6695
• Rouge2 Precision: 16.2843
• Rouge2 Recall: 12.9093
• Rouge2 Fmeasure: 13.9246
• Rougel Precision: 35.2648
• Rougel Recall: 28.6919
• Rougel Fmeasure: 30.7578
• Rougelsum Precision: 35.2646
• Rougelsum Recall: 28.6829
• Rougelsum Fmeasure: 30.7527
• Bleu-1: 23.9098
• Bleu-2: 14.7458
• Bleu-3: 9.684
• Bleu-4: 6.6217
• Meteor: 0.142

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1 Precision	Rouge1 Recall	Rouge1 Fmeasure	Rouge2 Precision	Rouge2 Recall	Rouge2 Fmeasure	Rougel Precision	Rougel Recall	Rougel Fmeasure	Rougelsum Precision	Rougelsum Recall	Rougelsum Fmeasure	Bleu-1	Bleu-2	Bleu-3	Bleu-4	Meteor
0.698	1.0	6769	0.5654	35.1173	28.5689	30.6164	14.7565	11.6885	12.6012	33.0241	26.9309	28.8245	33.0061	26.9075	28.807	22.6163	13.6841	8.8346	5.926	0.1314
0.6202	2.0	13538	0.5437	36.3795	29.5116	31.6675	15.5398	12.3022	13.2805	34.3036	27.8749	29.8881	34.2498	27.8384	29.8439	23.2744	14.1999	9.2715	6.2908	0.1364
0.5878	3.0	20307	0.5322	37.2522	30.1185	32.3701	16.0437	12.6396	13.6664	35.0062	28.3657	30.4487	34.9742	28.3319	30.4195	23.7569	14.5781	9.5429	6.52	0.1407
0.5761	4.0	27076	0.5309	37.5	30.4513	32.6723	16.2813	12.9079	13.9284	35.2662	28.6924	30.755	35.2509	28.6759	30.7444	23.9098	14.7458	9.684	6.6217	0.142

Framework versions

• Transformers 4.24.0
• Pytorch 1.12.1+cu113
• Datasets 2.6.1
• Tokenizers 0.13.2

Citation

@misc {tahsin_mayeesha_2023, author = { {Tahsin Mayeesha} }, title = { squad-bn-mt5-base2 (Revision 4ab9b63) }, year = 2023, url = { https://huggingface.co/Tahsin-Mayeesha/squad-bn-mt5-base2 }, doi = { 10.57967/hf/0940 }, publisher = { Hugging Face } }