karato-whisper-nqo

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

• Loss: 1.5661
• Wer: 0.7988
• Cer: 0.3860

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: 0.001
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• 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: 0.1
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.0854	0.1435	100	3.0519	1.5045	1.6379
2.4363	0.2869	200	2.4181	4.8672	2.1805
2.1652	0.4304	300	2.1447	5.4395	2.6315
1.8762	0.5739	400	1.8939	3.7282	1.8686
1.8097	0.7174	500	1.8239	2.8737	1.3935
1.7800	0.8608	600	1.7657	1.6866	0.7822
1.7366	1.0043	700	1.7525	1.4751	0.6785
1.7316	1.1478	800	1.7150	1.7459	0.8591
1.6995	1.2912	900	1.6929	1.6767	0.7688
1.7148	1.4347	1000	1.6759	2.2452	1.0648
1.6828	1.5782	1100	1.6645	1.4520	0.6944
1.6501	1.7217	1200	1.6489	1.1492	0.5792
1.6756	1.8651	1300	1.6447	1.3028	0.6159
1.5875	2.0086	1400	1.6483	1.0244	0.5125
1.6311	2.1521	1500	1.6297	1.2862	0.6047
1.6097	2.2956	1600	1.6225	1.1585	0.5608
1.6102	2.4390	1700	1.6231	1.1801	0.5428
1.6120	2.5825	1800	1.6178	1.0809	0.5105
1.5918	2.7260	1900	1.6107	1.0478	0.5045
1.5872	2.8694	2000	1.6055	1.1490	0.5313
1.5708	3.0129	2100	1.6012	1.0859	0.5261
1.5541	3.1564	2200	1.5989	1.1947	0.5589
1.5686	3.2999	2300	1.5961	1.0842	0.5187
1.5742	3.4433	2400	1.5956	1.2008	0.5329
1.5669	3.5868	2500	1.5853	1.0092	0.4676
1.5680	3.7303	2600	1.5875	1.1355	0.5293
1.5581	3.8737	2700	1.5795	1.0053	0.4986
1.5397	4.0172	2800	1.5820	0.9477	0.4239
1.5499	4.1607	2900	1.5777	1.0445	0.4709
1.5550	4.3042	3000	1.5800	0.9592	0.4397
1.5137	4.4476	3100	1.5766	0.9702	0.4651
1.5149	4.5911	3200	1.5745	0.8892	0.4060
1.5422	4.7346	3300	1.5657	0.9414	0.4317
1.5156	4.8780	3400	1.5646	1.0655	0.4735
1.4791	5.0215	3500	1.5701	0.9012	0.4137
1.5361	5.1650	3600	1.5706	0.9003	0.4262
1.4865	5.3085	3700	1.5646	0.9894	0.4515
1.5216	5.4519	3800	1.5627	1.0366	0.4696
1.5023	5.5954	3900	1.5594	1.0067	0.4537
1.4951	5.7389	4000	1.5584	0.9370	0.4320
1.4955	5.8824	4100	1.5600	0.8864	0.4108
1.4737	6.0258	4200	1.5595	0.9164	0.4150
1.4906	6.1693	4300	1.5594	0.9578	0.4243
1.4868	6.3128	4400	1.5589	0.8957	0.4072
1.4803	6.4562	4500	1.5617	0.8748	0.3913
1.4779	6.5997	4600	1.5539	0.8957	0.4005
1.4726	6.7432	4700	1.5536	0.8663	0.4053
1.4697	6.8867	4800	1.5512	0.9742	0.4276
1.4277	7.0301	4900	1.5587	0.8685	0.3974
1.4650	7.1736	5000	1.5575	0.8578	0.4091
1.4363	7.3171	5100	1.5576	0.8845	0.4182
1.4712	7.4605	5200	1.5567	0.8278	0.3889
1.4560	7.6040	5300	1.5549	0.8699	0.4035
1.4689	7.7475	5400	1.5508	0.8605	0.4047
1.4316	7.8910	5500	1.5519	0.8489	0.3986
1.4350	8.0344	5600	1.5617	0.8209	0.3916
1.4309	8.1779	5700	1.5633	0.8433	0.3907
1.4093	8.3214	5800	1.5592	0.8563	0.3921
1.4298	8.4648	5900	1.5651	0.8132	0.3926
1.4318	8.6083	6000	1.5576	0.8273	0.3947
1.4285	8.7518	6100	1.5523	0.8284	0.3966
1.4320	8.8953	6200	1.5472	0.8715	0.4047
1.4045	9.0387	6300	1.5614	0.8356	0.3946
1.4198	9.1822	6400	1.5707	0.7919	0.3847
1.3966	9.3257	6500	1.5633	0.8197	0.3935
1.4166	9.4692	6600	1.5629	0.8359	0.4044
1.3922	9.6126	6700	1.5622	0.8191	0.3910
1.3992	9.7561	6800	1.5635	0.8160	0.3925
1.4118	9.8996	6900	1.5661	0.7988	0.3860

Framework versions

• PEFT 0.18.1
• Transformers 5.3.0.dev0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2