whisper-large-v3-turbo-polyglot-lion

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4785
• Wer: 24.5798
• Cer: 11.1678

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 61875
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.4088	0.0162	1000	0.4994	34.7783	17.1210
0.3439	0.0323	2000	0.4525	41.2640	20.8228
0.418	0.0485	3000	0.4299	42.2343	20.5350
0.2897	0.0646	4000	0.4211	35.2814	15.9274
0.2222	0.0808	5000	0.4225	29.3957	13.0202
0.2104	0.0970	6000	0.4108	29.9707	13.1356
0.1934	0.1131	7000	0.4154	32.1049	15.2465
0.3047	0.1293	8000	0.3850	30.6065	14.1824
0.2644	0.1455	9000	0.3838	28.8179	13.2065
0.1836	0.1616	10000	0.3997	31.4719	15.0610
0.185	0.1778	11000	0.4163	29.4316	14.4171
0.1494	0.1939	12000	0.3826	29.0694	14.2108
0.1277	0.2101	13000	0.4084	28.4225	12.8984
0.117	0.2263	14000	0.3981	27.8613	12.4497
0.242	0.2424	15000	0.4118	26.9352	12.2379
0.1699	0.2586	16000	0.3818	26.1114	11.1309
0.1191	0.2747	17000	0.3833	27.8088	13.2176
0.1487	0.2909	18000	0.3916	26.7970	12.0083
0.1004	0.3071	19000	0.4037	31.1954	14.7147
0.0841	0.3232	20000	0.4048	26.0008	11.1771
0.0875	0.3394	21000	0.3925	27.0292	12.0558
0.0727	0.3556	22000	0.4144	26.8716	12.3126
0.1333	0.3717	23000	0.4026	25.5557	11.5201
0.091	0.3879	24000	0.4064	26.7693	12.3877
0.0908	0.4040	25000	0.4009	26.4790	11.8843
0.0804	1.0162	26000	0.4156	26.2385	11.9548
0.054	1.0323	27000	0.4205	26.0229	11.6793
0.1282	1.0485	28000	0.4019	26.2441	11.9102
0.0726	1.0646	29000	0.4040	26.1722	12.1284
0.0496	1.0808	30000	0.4223	25.1825	11.1750
0.0531	1.0970	31000	0.4093	25.9068	11.8168
0.0443	1.1131	32000	0.4364	26.1418	12.4255
0.0937	1.1293	33000	0.4330	25.5861	11.5647
0.0703	1.1455	34000	0.4180	26.5730	12.3618
0.04	1.1616	35000	0.4377	26.0727	12.0728
0.0536	1.1778	36000	0.4444	25.5170	12.4790
0.0367	1.1939	37000	0.4252	24.5881	11.6980
0.0346	1.2101	38000	0.4296	24.2812	10.5383
0.0336	1.2263	39000	0.4301	26.1694	12.6560
0.0948	1.2424	40000	0.4432	25.8377	11.2794
0.0411	1.2586	41000	0.4475	24.8397	11.8024
0.0251	1.2747	42000	0.4428	25.3262	12.5354
0.0432	1.2909	43000	0.4276	25.4202	12.2324
0.0288	1.3071	44000	0.4481	25.2571	12.6305
0.0216	1.3232	45000	0.4545	24.9392	11.7502
0.0242	1.3394	46000	0.4440	25.4755	13.1454
0.0184	1.3556	47000	0.4589	24.1181	10.6843
0.0403	1.3717	48000	0.4527	24.2149	10.6741
0.0191	1.3879	49000	0.4652	25.1880	11.4891
0.02	1.4040	50000	0.4579	23.7394	10.4636
0.0195	2.0162	51000	0.4696	24.3172	10.9025
0.0101	2.0323	52000	0.4759	24.6378	11.0961
0.0377	2.0485	53000	0.4617	25.7658	12.6250
0.0212	2.0646	54000	0.4592	24.1098	10.9182
0.0114	2.0808	55000	0.4715	23.8859	10.6160
0.0139	2.0970	56000	0.4606	24.3476	11.2973
0.0101	2.1131	57000	0.4779	24.0379	10.4254
0.0266	2.1293	58000	0.4752	23.9191	10.7200
0.0152	2.1455	59000	0.4749	24.3310	11.1215
0.0065	2.1616	60000	0.4764	24.1679	10.6572
0.0116	2.1778	61000	0.4785	24.5798	11.1678

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.1+cu128
• Datasets 3.5.0
• Tokenizers 0.22.1