sinhala-asr-correction

This model is a fine-tuned version of facebook/mbart-large-50 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8496
• Bleu: 57.57
• Cer: 0.0789

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: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Cer
0.7743	1.0	1765	0.7391	48.49	0.0912
0.5622	2.0	3530	0.6628	51.18	0.0905
0.3959	3.0	5295	0.6445	52.41	0.0869
0.2465	4.0	7060	0.6916	54.07	0.0847
0.1489	5.0	8825	0.7436	55.23	0.0851
0.0844	6.0	10590	0.7834	56.07	0.0832
0.0483	7.0	12355	0.8079	56.27	0.0826
0.0284	8.0	14120	0.8208	57.41	0.0816
0.0173	9.0	15885	0.8438	57.33	0.0801
0.0134	10.0	17650	0.8496	57.57	0.0789

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2