superb-wav2vec2

This model is a fine-tuned version of vasista22/ccc-wav2vec2-base-SUPERB on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0003
• Wer: 0.0233

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: 0.0004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 132
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.0211	0.4082	50	2.0267	0.9861
1.9496	0.8163	100	1.7685	0.9849
1.7178	1.2245	150	1.4738	0.8240
1.3801	1.6327	200	1.1281	0.8227
1.189	2.0408	250	0.8568	0.5723
0.9318	2.4490	300	0.6622	0.5615
0.7042	2.8571	350	0.3612	0.3023
0.5805	3.2653	400	0.4220	0.4606
0.4229	3.6735	450	0.1465	0.1417
0.3913	4.0816	500	0.1350	0.1688
0.2645	4.4898	550	0.1030	0.1421
0.2809	4.8980	600	0.0867	0.0977
0.2344	5.3061	650	0.0901	0.1367
0.1703	5.7143	700	0.0659	0.1246
0.1718	6.1224	750	0.0432	0.0545
0.1442	6.5306	800	0.0636	0.0824
0.1494	6.9388	850	0.0431	0.0448
0.1492	7.3469	900	0.0328	0.0478
0.1185	7.7551	950	0.0376	0.0621
0.107	8.1633	1000	0.0249	0.0241
0.1159	8.5714	1050	0.0350	0.0396
0.1015	8.9796	1100	0.0232	0.0334
0.1203	9.3878	1150	0.0341	0.0780
0.0835	9.7959	1200	0.0178	0.0458
0.1239	10.2041	1250	0.0231	0.0543
0.0859	10.6122	1300	0.0163	0.0289
0.0732	11.0204	1350	0.0309	0.0494
0.063	11.4286	1400	0.0168	0.0963
0.0693	11.8367	1450	0.0268	0.0619
0.0649	12.2449	1500	0.0328	0.0687
0.063	12.6531	1550	0.0173	0.0438
0.0574	13.0612	1600	0.0118	0.0506
0.0438	13.4694	1650	0.0101	0.0510
0.0556	13.8776	1700	0.0064	0.0291
0.0536	14.2857	1750	0.0098	0.0225
0.047	14.6939	1800	0.0157	0.0251
0.0588	15.1020	1850	0.0097	0.0291
0.0397	15.5102	1900	0.0113	0.0541
0.0375	15.9184	1950	0.0173	0.0531
0.0411	16.3265	2000	0.0079	0.0394
0.0382	16.7347	2050	0.0056	0.0340
0.0448	17.1429	2100	0.0064	0.0287
0.0359	17.5510	2150	0.0053	0.0261
0.032	17.9592	2200	0.0091	0.0400
0.0295	18.3673	2250	0.0018	0.0275
0.03	18.7755	2300	0.0034	0.0259
0.0246	19.1837	2350	0.0280	0.0368
0.0465	19.5918	2400	0.0099	0.0297
0.0264	20.0	2450	0.0063	0.0111
0.025	20.4082	2500	0.0015	0.0370
0.04	20.8163	2550	0.0020	0.0344
0.0203	21.2245	2600	0.0055	0.0356
0.0241	21.6327	2650	0.0024	0.0299
0.0465	22.0408	2700	0.0022	0.0392
0.0283	22.4490	2750	0.0026	0.0149
0.0134	22.8571	2800	0.0015	0.0177
0.0177	23.2653	2850	0.0041	0.0177
0.0288	23.6735	2900	0.0011	0.0147
0.0216	24.0816	2950	0.0034	0.0287
0.0147	24.4898	3000	0.0046	0.0155
0.0118	24.8980	3050	0.0021	0.0235
0.0113	25.3061	3100	0.0012	0.0261
0.0135	25.7143	3150	0.0006	0.0261
0.0118	26.1224	3200	0.0008	0.0287
0.0083	26.5306	3250	0.0004	0.0257
0.0148	26.9388	3300	0.0006	0.0261
0.0081	27.3469	3350	0.0005	0.0263
0.0192	27.7551	3400	0.0004	0.0237
0.0096	28.1633	3450	0.0004	0.0231
0.0083	28.5714	3500	0.0003	0.0215
0.0056	28.9796	3550	0.0004	0.0233
0.0082	29.3878	3600	0.0003	0.0233
0.0102	29.7959	3650	0.0003	0.0233

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.0
• Datasets 2.21.0
• Tokenizers 0.19.1