mt5-small-nepali-v3-final

This model is a fine-tuned version of aku47/mt5-small-nepali-v2-final on an unknown dataset. It achieves the following results on the evaluation set:

• Loss : 1.3185
• rouge1 : 44.6707
• rouge2 : 30.6280
• rougeL : 39.3135
• rougeLsum : 39.2927
• bertscore_precision : 79.9863
• bertscore_recall : 78.9514
• vertscore_f1 : 79.4001

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-06
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Bertscore Precision	Bertscore Recall	Bertscore F1	Gen Len
2.9619	1.0	487	1.3398	0.1928	0.1021	0.1722	0.1724	0.7673	0.6588	0.7084	21.0
2.3409	2.0	974	1.3373	0.1907	0.1148	0.1801	0.1804	0.7757	0.6572	0.7108	21.0
2.092	3.0	1461	1.3318	0.1946	0.1231	0.1853	0.1857	0.7788	0.6588	0.713	21.0
2.0301	4.0	1948	1.3215	0.1993	0.1287	0.1898	0.1903	0.7814	0.6597	0.7146	21.0
2.0253	5.0	2435	1.3185	0.2018	0.1312	0.192	0.1926	0.7826	0.6603	0.7155	21.0

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1