whisper-small-fft-commonvoice-mongolian-ver_0.2

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

• Loss: 0.2859
• Wer: 0.3623
• Cer: 0.1281

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: 3.5e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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: 500
• training_steps: 8000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
16.0138	0.1098	200	2.0730	0.9644	0.8035
8.9194	0.2197	400	1.2396	0.8294	0.3440
6.7664	0.3295	600	0.7964	0.7354	0.2902
5.4246	0.4393	800	0.6543	0.6679	0.2480
5.0296	0.5491	1000	0.5834	0.6173	0.2299
4.7781	0.6590	1200	0.5336	0.5890	0.2146
4.4746	0.7688	1400	0.4961	0.5539	0.2012
4.2903	0.8786	1600	0.4699	0.5200	0.1835
4.2277	0.9885	1800	0.4449	0.5047	0.1814
3.8068	1.0983	2000	0.4255	0.4921	0.1767
3.5961	1.2081	2200	0.4083	0.4728	0.1656
3.6018	1.3180	2400	0.3957	0.4563	0.1623
3.4267	1.4278	2600	0.3825	0.4472	0.1564
3.3365	1.5376	2800	0.3726	0.4394	0.1554
3.2877	1.6474	3000	0.3630	0.4322	0.1523
3.1749	1.7573	3200	0.3540	0.4241	0.1492
3.1345	1.8671	3400	0.3461	0.4157	0.1470
2.9984	1.9769	3600	0.3397	0.4062	0.1423
2.7815	2.0868	3800	0.3327	0.4027	0.1434
2.6561	2.1966	4000	0.3278	0.3998	0.1423
2.6535	2.3064	4200	0.3254	0.3962	0.1400
2.5770	2.4163	4400	0.3191	0.3883	0.1375
2.5749	2.5261	4600	0.3152	0.3849	0.1353
2.4985	2.6359	4800	0.3095	0.3842	0.1354
2.4846	2.7457	5000	0.3072	0.3796	0.1349
2.4470	2.8556	5200	0.3032	0.3772	0.1333
2.4002	2.9654	5400	0.2993	0.3772	0.1346
2.2121	3.0752	5600	0.2995	0.3733	0.1322
2.1778	3.1851	5800	0.2983	0.3727	0.1316
2.1431	3.2949	6000	0.2953	0.3698	0.1314
2.0874	3.4047	6200	0.2949	0.3658	0.1294
2.1652	3.5146	6400	0.2917	0.3691	0.1315
2.1117	3.6244	6600	0.2906	0.3613	0.1261
2.1010	3.7342	6800	0.2894	0.3657	0.1300
2.0794	3.8440	7000	0.2878	0.3609	0.1272
2.0773	3.9539	7200	0.2864	0.3604	0.1272
1.9416	4.0637	7400	0.2864	0.3634	0.1284
1.9862	4.1735	7600	0.2864	0.3632	0.1278
1.9737	4.2834	7800	0.2864	0.3614	0.1278
1.9325	4.3932	8000	0.2859	0.3623	0.1281

Framework versions

• Transformers 5.5.0
• Pytorch 2.11.0+cu128
• Datasets 4.8.4
• Tokenizers 0.22.2