Whisper-WOLOF-40-hours-Kallaama-dataset

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: 1.2031
• Wer: 31.4362
• Cer: 17.5733

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: 32
• eval_batch_size: 8
• seed: 42
• 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: 0.1
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.6381	0.8865	500	1.4538	78.5863	46.2285
1.146	1.7730	1000	0.9648	65.8870	39.5269
0.8223	2.6596	1500	0.7969	51.5679	30.6941
0.6542	3.5461	2000	0.7221	51.9664	31.0189
0.538	4.4326	2500	0.6842	43.1480	25.9115
0.4373	5.3191	3000	0.6807	42.8274	25.0277
0.3507	6.2057	3500	0.6903	38.5395	22.5312
0.2682	7.0922	4000	0.7086	34.8233	19.8022
0.1924	7.9787	4500	0.7283	36.1227	21.9131
0.1324	8.8652	5000	0.7627	39.5530	24.4465
0.0964	9.7518	5500	0.7997	36.9196	21.1344
0.0632	10.6383	6000	0.8383	34.6327	19.2284
0.0444	11.5248	6500	0.8721	36.6078	21.0329
0.0298	12.4113	7000	0.8992	33.0215	18.5752
0.0204	13.2979	7500	0.9203	32.8742	18.1360
0.0158	14.1844	8000	0.9349	32.7096	18.4977
0.013	15.0709	8500	0.9601	34.0437	18.9774
0.0108	15.9574	9000	0.9658	33.1341	18.5401
0.0095	16.8440	9500	0.9900	33.7318	18.9424
0.0084	17.7305	10000	1.0027	33.8444	19.5254
0.0075	18.6170	10500	1.0261	32.5710	18.4405
0.0065	19.5035	11000	1.0283	32.6923	18.2670
0.0057	20.3901	11500	1.0342	32.2852	18.2818
0.0054	21.2766	12000	1.0449	32.4324	18.3870
0.0053	22.1631	12500	1.0484	32.4411	18.4423
0.0046	23.0496	13000	1.0533	32.2419	18.1508
0.004	23.9362	13500	1.0694	32.4584	17.9995
0.0035	24.8227	14000	1.0747	32.1812	17.8998
0.0033	25.7092	14500	1.0827	32.6836	18.3371
0.0032	26.5957	15000	1.0824	32.7616	18.4940
0.0026	27.4823	15500	1.0965	31.9820	17.8482
0.0025	28.3688	16000	1.1104	31.6875	17.6120
0.0022	29.2553	16500	1.1076	33.1947	18.6453
0.0021	30.1418	17000	1.1103	32.6836	18.3519
0.0018	31.0284	17500	1.1136	31.4103	17.1415
0.0017	31.9149	18000	1.1142	32.0599	18.0844
0.0018	32.8014	18500	1.1220	31.5142	17.7209
0.0014	33.6879	19000	1.1192	31.8867	17.7873
0.001	34.5745	19500	1.1256	32.0773	18.1305
0.0008	35.4610	20000	1.1326	31.8347	17.5991
0.0009	36.3475	20500	1.1430	31.3756	17.3426
0.0007	37.2340	21000	1.1430	31.6268	17.7393
0.0008	38.1206	21500	1.1548	32.2765	18.3814
0.0008	39.0071	22000	1.1528	31.0638	17.1655
0.0005	39.8936	22500	1.1551	31.3669	17.4589
0.0006	40.7801	23000	1.1581	31.3150	17.3611
0.0005	41.6667	23500	1.1650	31.9387	17.9349
0.0003	42.5532	24000	1.1804	30.6566	16.8278
0.0003	43.4397	24500	1.1770	31.7827	17.7818
0.0002	44.3262	25000	1.1848	31.8520	18.0844
0.0003	45.2128	25500	1.1853	31.9040	17.9534
0.0002	46.0993	26000	1.1888	32.1552	18.2836
0.0002	46.9858	26500	1.1925	31.4709	17.6637
0.0002	47.8723	27000	1.1984	31.8174	17.8260
0.0001	48.7589	27500	1.2009	31.4016	17.5972
0.0001	49.6454	28000	1.2031	31.4362	17.5733

Framework versions

• Transformers 4.46.0
• Pytorch 2.1.0+cu118
• Datasets 3.0.2
• Tokenizers 0.20.1