legendary2910/Mnong-ASR-v1

Whisper Medium Mnong

This model is a fine-tuned version of openai/whisper-medium on the MnongAudio-v2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2773
• Wer: 16.8874

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1592	0.2915	200	2.0293	117.3459
1.3786	0.5831	400	1.3853	90.0408
0.9096	0.8746	600	1.0002	77.8146
0.5869	1.1662	800	0.7615	64.7733
0.4996	1.4577	1000	0.5799	52.1141
0.3741	1.7493	1200	0.4811	60.9781
0.1899	2.0408	1400	0.4078	35.3031
0.1792	2.3324	1600	0.3690	34.2588
0.1514	2.6239	1800	0.3361	31.5079
0.1758	2.9155	2000	0.3069	30.9730
0.0619	3.2070	2200	0.3031	28.2731
0.047	3.4985	2400	0.2952	22.1600
0.0472	3.7901	2600	0.2914	24.8344
0.0246	4.0816	2800	0.2799	20.0458
0.0255	4.3732	3000	0.2849	23.4590
0.0246	4.6647	3200	0.2773	19.5619
0.0235	4.9563	3400	0.2736	20.4279
0.0088	5.2478	3600	0.2795	19.9440
0.0076	5.5394	3800	0.2786	17.0657
0.0057	5.8309	4000	0.2773	16.8874

Framework versions

• Transformers 4.42.4
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1