myatsu/mms-1b-burmese-v1

MMS-1B Burmese ASR - SLR80 Fine-Tuned

This model is a fine-tuned version of facebook/mms-1b-all on the Myanmar Speech Dataset (OpenSLR-80) dataset. It achieves the following results on the evaluation set:

• Loss: 0.3057
• Wer: 0.9486
• Cer: 0.1060

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
12.2228	0.7027	100	0.5757	1.0	0.2000
11.0420	1.4005	200	0.5379	1.0	0.1918
9.6407	2.0984	300	0.4368	0.9921	0.1592
8.5662	2.8011	400	0.4035	0.9842	0.1474
6.6940	3.4989	500	0.3860	1.0	0.1394
4.9239	4.1968	600	0.3418	0.9921	0.1261
4.7416	4.8994	700	0.3377	0.9723	0.1231
4.0854	5.5973	800	0.3226	0.9644	0.1128
3.4215	6.2951	900	0.3100	0.9526	0.1072
3.1916	6.9978	1000	0.3057	0.9486	0.1059

Framework versions

• Transformers 5.0.0.dev0
• Pytorch 2.9.0+cu126
• Datasets 4.4.2
• Tokenizers 0.22.1