whisper-large-v3-turbo-latvian-latvia-0.06-cer-filtered

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.4949
• Model Preparation Time: 0.0064
• Wer Ortho: 35.0098
• Wer: 11.0794

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: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• optimizer: Use 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_ratio: 0.06
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer Ortho	Wer
0.2464	0.0837	32	0.4843	0.0064	38.5584	17.2289
0.2141	0.1673	64	0.4760	0.0064	38.4149	16.7942
0.2035	0.2510	96	0.4884	0.0064	38.0561	16.1650
0.1864	0.3346	128	0.4361	0.0064	37.3190	15.5423
0.1742	0.4183	160	0.4530	0.0064	37.4364	15.2763
0.1534	0.5020	192	0.4413	0.0064	37.2342	14.4266
0.1608	0.5856	224	0.4777	0.0064	36.2427	14.0893
0.1655	0.6693	256	0.4263	0.0064	36.1057	13.6611
0.1555	0.7529	288	0.4399	0.0064	35.9817	13.3433
0.1507	0.8366	320	0.4227	0.0064	35.6882	13.0773
0.1495	0.9203	352	0.4559	0.0064	35.9295	13.0384
0.138	1.0026	384	0.4431	0.0064	35.6295	12.7530
0.1165	1.0863	416	0.4746	0.0064	35.9361	12.3832
0.1211	1.1699	448	0.4599	0.0064	35.6295	12.1757
0.1177	1.2536	480	0.4726	0.0064	35.8317	12.3703
0.1208	1.3373	512	0.4590	0.0064	35.7991	12.3508
0.118	1.4209	544	0.4564	0.0064	35.7078	12.2146
0.1179	1.5046	576	0.4346	0.0064	35.4338	11.8773
0.1206	1.5882	608	0.4512	0.0064	35.5447	11.7475
0.1167	1.6719	640	0.4653	0.0064	35.3164	11.6697
0.1065	1.7556	672	0.4559	0.0064	35.2120	11.5789
0.1026	1.8392	704	0.4734	0.0064	34.9772	11.5335
0.1128	1.9229	736	0.4723	0.0064	35.0294	11.5400
0.1142	2.0052	768	0.4784	0.0064	34.9772	11.5335
0.0941	2.0889	800	0.4937	0.0064	35.2577	11.4946
0.0874	2.1725	832	0.4896	0.0064	35.0424	11.4232
0.0842	2.2562	864	0.4933	0.0064	35.1076	11.3324
0.082	2.3399	896	0.4811	0.0064	34.6510	11.2156
0.0864	2.4235	928	0.4802	0.0064	34.5597	11.0924
0.0755	2.5072	960	0.4954	0.0064	34.8206	11.3389
0.0833	2.5908	992	0.4900	0.0064	34.9967	11.2026
0.0956	2.6745	1024	0.4928	0.0064	34.9511	11.1572
0.0801	2.7582	1056	0.4872	0.0064	34.7032	11.1313
0.0782	2.8418	1088	0.4949	0.0064	35.0098	11.0794
0.0851	2.9255	1120	0.4979	0.0064	34.9119	11.1508

Framework versions

• Transformers 4.51.3
• Pytorch 2.5.1
• Datasets 3.6.0
• Tokenizers 0.21.1