whisper-small-mongolian-ver_0.1

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

• Cer: 0.2079
• Loss: 1.0889
• Wer: 0.5708

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 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
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
16.2963	0.7835	200	0.7531	2.0496	0.9645
8.2355	1.5642	400	0.3457	1.2264	0.8376
5.9008	2.3448	600	0.2816	0.8037	0.7203
4.6743	3.1254	800	0.2440	0.6909	0.6671
4.0781	3.9089	1000	0.2323	0.6291	0.6298
3.5529	4.6895	1200	0.2233	0.6075	0.6044
3.2198	5.4701	1400	0.2090	0.5985	0.5786
2.9363	6.2507	1600	0.2029	0.5972	0.5667
2.6100	7.0313	1800	0.2032	0.5931	0.5597
2.3090	7.8149	2000	0.2022	0.6008	0.5557
1.9680	8.5955	2200	0.2021	0.6182	0.5560
1.8022	9.3761	2400	0.2020	0.6352	0.5634
1.5629	10.1567	2600	0.2061	0.6515	0.5627
1.3567	10.9403	2800	0.2009	0.6567	0.5574
1.0980	11.7209	3000	0.2050	0.6807	0.5653
0.9251	12.5015	3200	0.2096	0.6964	0.5718
0.8264	13.2821	3400	0.2083	0.7131	0.5709
0.6794	14.0627	3600	0.2084	0.7299	0.5700
0.5547	14.8462	3800	0.2072	0.7402	0.5632
0.4671	15.6268	4000	0.2076	0.7545	0.5633
0.3914	16.4074	4200	0.2044	0.7683	0.5650
0.3439	17.1881	4400	0.2093	0.7837	0.5686
0.2910	17.9716	4600	0.2155	0.7909	0.5776
0.2448	18.7522	4800	0.2074	0.8097	0.5704
0.2124	19.5328	5000	0.2080	0.8215	0.5685
0.1952	20.3134	5200	0.2055	0.8352	0.5670
0.1704	21.0940	5400	0.2036	0.8472	0.5643
0.1527	21.8776	5600	0.2042	0.8652	0.5632
0.1413	22.6582	5800	0.2089	0.8803	0.5689
0.1329	23.4388	6000	0.2060	0.8928	0.5667
0.1227	24.2194	6200	0.2058	0.9047	0.5648
0.1183	25.0	6400	0.2092	0.9146	0.5687
0.1129	25.7835	6600	0.2054	0.9257	0.5646
0.1103	26.5642	6800	0.2055	0.9421	0.5662
0.1088	27.3448	7000	0.2034	0.9395	0.5633
0.1066	28.1254	7200	0.2106	0.9415	0.5690
0.1055	28.9089	7400	0.2010	0.9535	0.5605
0.1006	29.6895	7600	0.2088	0.9612	0.5649
0.1000	30.4701	7800	0.2088	0.9723	0.5653
0.0995	31.2507	8000	0.2023	0.9783	0.5625
0.0985	32.0313	8200	0.2051	0.9885	0.5639
0.0973	32.8149	8400	0.2030	0.9971	0.5640
0.0960	33.5955	8600	0.2064	1.0033	0.5639
0.0957	34.3761	8800	0.2066	1.0103	0.5677
0.0947	35.1567	9000	0.2078	1.0181	0.5677
0.0942	35.9403	9200	0.2100	1.0237	0.5715
0.0932	36.7209	9400	0.2071	1.0276	0.5677
0.0937	37.5015	9600	0.2133	1.0345	0.5739
0.0931	38.2821	9800	0.2104	1.0406	0.5738
0.0926	39.0627	10000	0.2072	1.0452	0.5680
0.0933	39.8462	10200	0.2089	1.0478	0.5706
0.0914	40.6268	10400	0.2105	1.0540	0.5714
0.0925	41.4074	10600	0.2082	1.0606	0.5690
0.0921	42.1881	10800	0.2080	1.0648	0.5706
0.0918	42.9716	11000	0.2069	1.0666	0.5710
0.0913	43.7522	11200	0.2087	1.0716	0.5679
0.0905	44.5328	11400	0.2077	1.0732	0.5699
0.0916	45.3134	11600	0.2082	1.0800	0.5704
0.0911	46.0940	11800	0.2080	1.0811	0.5704
0.0910	46.8776	12000	0.2083	1.0833	0.5706
0.0898	47.6582	12200	0.2077	1.0862	0.5706
0.0906	48.4388	12400	0.2069	1.0878	0.5718
0.0912	49.2194	12600	0.2079	1.0886	0.5711
0.0904	50.0	12800	0.2079	1.0889	0.5708

Framework versions

• Transformers 5.4.0
• Pytorch 2.6.0+cu124
• Datasets 2.19.0
• Tokenizers 0.22.2