legendary2910/Mnong-ASR-v2

Whisper Small Mnong

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

• Loss: 0.3726
• Wer: 37.8757

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
3.2302	0.2915	200	3.1116	239.5568
1.7352	0.5831	400	1.7781	96.5359
1.2762	0.8746	600	1.3745	100.9170
0.9605	1.1662	800	1.1114	89.1747
0.8392	1.4577	1000	0.9010	81.3551
0.6779	1.7493	1200	0.7770	64.3912
0.3956	2.0408	1400	0.6635	63.0922
0.35	2.3324	1600	0.6001	55.6037
0.3225	2.6239	1800	0.5402	59.2970
0.326	2.9155	2000	0.4830	48.0387
0.143	3.2070	2200	0.4697	41.7983
0.119	3.4985	2400	0.4404	41.5181
0.1123	3.7901	2600	0.4205	44.4727
0.0709	4.0816	2800	0.4009	38.9710
0.0764	4.3732	3000	0.3959	39.8370
0.0587	4.6647	3200	0.3800	35.8889
0.0627	4.9563	3400	0.3771	37.6465
0.0326	5.2478	3600	0.3754	35.3795
0.0355	5.5394	3800	0.3751	39.4549
0.0331	5.8309	4000	0.3726	37.8757

Framework versions

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