whisper-small-Kurdish-Sorani-10

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1166
• Wer Ortho: 14.4007
• Wer: 13.1989

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: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.2207	0.0992	1000	0.2857	48.3163	44.5546
0.1701	0.1984	2000	0.2396	41.9819	38.2653
0.1551	0.2976	3000	0.2099	37.3690	33.9086
0.1213	0.3968	4000	0.1918	34.8926	31.6996
0.1205	0.4960	5000	0.1757	32.6973	29.4823
0.1126	0.5952	6000	0.1654	31.8523	28.7945
0.1229	0.6944	7000	0.1520	29.2376	26.4927
0.0966	0.7937	8000	0.1459	28.1116	25.5538
0.0805	0.8929	9000	0.1345	26.0589	23.7183
0.0829	0.9921	10000	0.1290	25.4676	23.3069
0.0503	1.0913	11000	0.1261	24.1885	21.9946
0.0363	1.1905	12000	0.1212	23.0877	21.0642
0.0562	1.2897	13000	0.1177	22.5090	20.7266
0.0382	1.3889	14000	0.1152	21.6053	19.8785
0.0457	1.4881	15000	0.1143	21.0224	19.4502
0.0394	1.5873	16000	0.1072	20.3892	18.8130
0.0427	1.6865	17000	0.1066	19.8482	18.2814
0.03	1.7857	18000	0.1033	19.0619	17.5957
0.0311	1.8849	19000	0.1018	18.7390	17.2391
0.0308	1.9841	20000	0.1004	18.8753	17.3172
0.0297	2.0833	21000	0.1034	18.1309	16.7623
0.0158	2.1825	22000	0.1052	18.5042	17.1463
0.0157	2.2817	23000	0.1039	17.8290	16.4374
0.0367	2.3810	24000	0.1022	18.0953	16.8129
0.0144	2.4802	25000	0.1041	17.3551	16.0724
0.01	2.5794	26000	0.1051	17.3132	15.9880
0.0116	2.6786	27000	0.1046	16.8561	15.4711
0.0149	2.7778	28000	0.1011	16.9861	15.5914
0.02	2.8770	29000	0.1008	16.4367	15.1357
0.0122	2.9762	30000	0.1002	16.1914	14.9352
0.004	3.0754	31000	0.1057	15.6861	14.3403
0.0055	3.1746	32000	0.1067	15.7783	14.4795
0.0045	3.2738	33000	0.1089	15.7133	14.3761
0.0084	3.3730	34000	0.1072	15.7196	14.4500
0.0046	3.4722	35000	0.1087	15.7825	14.4837
0.0032	3.5714	36000	0.1094	15.3757	14.1567
0.0085	3.6706	37000	0.1071	15.4303	14.1989
0.0064	3.7698	38000	0.1106	15.2688	14.0280
0.0037	3.8690	39000	0.1086	14.9836	13.7263
0.0123	3.9683	40000	0.1109	14.7886	13.5639
0.0021	4.0675	41000	0.1135	14.7362	13.4900
0.0017	4.1667	42000	0.1142	14.5685	13.3402
0.0019	4.2659	43000	0.1144	14.6964	13.4141
0.0013	4.3651	44000	0.1156	14.6796	13.4225
0.0051	4.4643	45000	0.1155	14.5769	13.3381
0.001	4.5635	46000	0.1162	14.4846	13.2727
0.0008	4.6627	47000	0.1170	14.5119	13.3086
0.0045	4.7619	48000	0.1149	14.6083	13.4098
0.0012	4.8611	49000	0.1164	14.3609	13.1672
0.0007	4.9603	50000	0.1166	14.4007	13.1989

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu118
• Datasets 2.21.0
• Tokenizers 0.19.1