mt5-base-finetuned-urdu

This model is a fine-tuned version of google/mt5-base on Urdu subset the xlsum dataset. It achieves the following results on the evaluation set:

• Loss: 2.8954
• Rouge-1: 28.84
• Rouge-2: 13.87
• Rouge-l: 25.63
• Gen Len: 19.0
• Bertscore: 71.31

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: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5
• label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge-1	Rouge-2	Rouge-l	Gen Len	Bertscore
3.6205	1.0	2114	3.0871	26.45	11.4	23.26	19.0	70.76
3.2169	2.0	4228	2.9830	27.19	11.91	23.95	19.0	70.92
3.0787	3.0	6342	2.9284	27.9	12.57	24.62	18.99	71.13
2.9874	4.0	8456	2.9049	28.28	12.91	24.99	18.99	71.28
2.9232	5.0	10570	2.8954	28.65	13.17	25.32	18.99	71.39

Framework versions

• Transformers 4.18.0
• Pytorch 1.11.0+cu113
• Datasets 2.1.0
• Tokenizers 0.12.1