mT5_multilingual_XLSum-finetune-ar-xlsum

This model is a fine-tuned version of csebuetnlp/mT5_multilingual_XLSum on the xlsum dataset. It achieves the following results on the evaluation set:

• Loss: 3.2497
• Rouge-1: 32.52
• Rouge-2: 14.71
• Rouge-l: 27.88
• Gen Len: 41.45
• Bertscore: 74.65

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: 0.0005
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• num_epochs: 8
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge-1	Rouge-2	Rouge-l	Gen Len	Bertscore
3.5465	1.0	585	3.3215	30.09	13.23	26.07	36.31	73.97
3.3564	2.0	1170	3.2547	31.29	13.93	26.75	41.68	74.22
3.2185	3.0	1755	3.2421	31.78	14.1	27.07	41.64	74.4
3.1145	4.0	2340	3.2241	31.98	14.38	27.51	40.29	74.46
3.031	5.0	2925	3.2313	32.3	14.67	27.83	39.81	74.61
2.9627	6.0	3510	3.2348	32.39	14.65	27.76	40.02	74.6
2.9088	7.0	4095	3.2439	32.5	14.66	27.81	41.2	74.65
2.8649	8.0	4680	3.2497	32.52	14.71	27.88	41.45	74.65

Framework versions

• Transformers 4.19.4
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1