whisper-medium-r22-e_v231109

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3784
• Wer: 100.0

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5
• training_steps: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.0027	0.06	10	3.8236	29.3631
2.668	0.12	20	1.8668	27.7539
1.5247	0.18	30	1.0451	25.8273
0.7177	0.24	40	0.3820	100.0
0.342	0.3	50	0.3398	100.0
0.331	0.36	60	0.3243	100.0340
0.3139	0.42	70	0.3175	100.0227
0.291	0.48	80	0.2983	100.0340
0.3178	0.54	90	0.2907	100.0340
0.2516	0.6	100	0.2933	100.0567
0.3004	0.66	110	0.2860	100.0907
0.2923	0.72	120	0.2962	100.1587
0.3067	0.78	130	0.2887	100.0340
0.2967	0.84	140	0.2802	100.0
0.3059	0.9	150	0.2734	100.0
0.2465	0.96	160	0.2686	100.0
0.1953	1.02	170	0.2677	100.0793
0.1611	1.08	180	0.2665	100.0453
0.1548	1.14	190	0.2644	100.0
0.1379	1.2	200	0.2781	100.0
0.1593	1.27	210	0.2765	100.0
0.1266	1.33	220	0.2805	100.0
0.1407	1.39	230	0.2669	100.0567
0.1301	1.45	240	0.2708	100.0793
0.1546	1.51	250	0.2713	100.0793
0.1447	1.57	260	0.2723	100.0793
0.1762	1.63	270	0.2689	100.0
0.148	1.69	280	0.2693	100.0680
0.1468	1.75	290	0.2682	100.0340
0.1747	1.81	300	0.2688	100.0340
0.106	1.87	310	0.2606	100.0
0.1517	1.93	320	0.2606	100.0
0.143	1.99	330	0.2644	100.0
0.085	2.05	340	0.2644	100.0
0.0733	2.11	350	0.2840	100.0
0.0606	2.17	360	0.2879	100.0
0.071	2.23	370	0.2851	100.0
0.0518	2.29	380	0.2975	100.0
0.068	2.35	390	0.2936	100.0
0.0553	2.41	400	0.3062	100.0
0.049	2.47	410	0.3019	100.0
0.0621	2.53	420	0.3021	100.0
0.0593	2.59	430	0.2941	100.0
0.0604	2.65	440	0.2960	100.0
0.0711	2.71	450	0.2996	100.0
0.0643	2.77	460	0.2907	100.0
0.0554	2.83	470	0.2902	100.0
0.0595	2.89	480	0.2992	100.0
0.0693	2.95	490	0.2936	99.8527
0.0411	3.01	500	0.2937	100.0
0.0192	3.07	510	0.3174	100.0
0.0105	3.13	520	0.3468	100.0
0.0339	3.19	530	0.3439	100.0
0.0222	3.25	540	0.3571	100.0
0.0372	3.31	550	0.3393	100.0
0.0219	3.37	560	0.3468	100.0
0.0223	3.43	570	0.3341	100.0
0.0239	3.49	580	0.3393	100.0
0.0322	3.55	590	0.3378	100.0
0.0299	3.61	600	0.3296	100.0
0.0223	3.67	610	0.3367	100.0
0.0234	3.73	620	0.3345	100.0
0.0191	3.8	630	0.3395	100.0
0.0207	3.86	640	0.3439	100.0
0.0258	3.92	650	0.3440	100.0
0.0209	3.98	660	0.3442	100.0
0.0164	4.04	670	0.3551	100.0
0.0067	4.1	680	0.3559	100.0
0.0094	4.16	690	0.3628	100.0
0.0096	4.22	700	0.3661	100.0
0.0073	4.28	710	0.3682	100.0
0.0106	4.34	720	0.3717	100.0
0.0067	4.4	730	0.3749	100.0
0.005	4.46	740	0.3785	100.0
0.0101	4.52	750	0.3803	100.0
0.0084	4.58	760	0.3784	100.0
0.0079	4.64	770	0.3770	100.0
0.0038	4.7	780	0.3772	100.0
0.0057	4.76	790	0.3780	100.0
0.0103	4.82	800	0.3784	100.0

Framework versions

• Transformers 4.36.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.14.7.dev0
• Tokenizers 0.14.1