exp20-M04-both

This model is a fine-tuned version of yongjian/wav2vec2-large-a on the None dataset. It achieves the following results on the evaluation set:

• Loss: 3.3358
• Wer: 1.1374

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: 0.0001
• train_batch_size: 4
• 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: 1000
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
37.2423	0.34	500	3.2577	1.0
3.1334	0.68	1000	3.0084	1.0
2.9616	1.02	1500	2.9946	1.0
2.8067	1.36	2000	2.6228	1.3130
2.732	1.7	2500	2.5059	1.5013
2.3673	2.04	3000	2.1828	1.4911
2.1378	2.38	3500	2.2066	1.4911
1.9853	2.72	4000	1.9877	1.4580
1.8574	3.06	4500	1.8850	1.4656
1.7085	3.4	5000	1.9121	1.4606
1.6161	3.74	5500	2.1036	1.4326
1.5304	4.08	6000	1.9807	1.4478
1.3531	4.42	6500	2.0211	1.4656
1.3269	4.77	7000	1.9231	1.3893
1.2312	5.11	7500	2.2652	1.4097
1.1161	5.45	8000	1.9543	1.4529
1.0305	5.79	8500	2.1463	1.4071
0.9403	6.13	9000	3.7872	1.4071
0.8723	6.47	9500	2.8466	1.4326
0.8752	6.81	10000	2.2215	1.3766
0.7774	7.15	10500	2.0462	1.3257
0.74	7.49	11000	2.1928	1.3333
0.7371	7.83	11500	2.8058	1.3410
0.7075	8.17	12000	2.3100	1.3308
0.6746	8.51	12500	2.6284	1.2875
0.6233	8.85	13000	2.2268	1.3003
0.7172	9.19	13500	2.1980	1.2926
0.5697	9.53	14000	2.1950	1.2468
0.5691	9.87	14500	2.1819	1.2316
0.5062	10.21	15000	2.1426	1.2621
0.4818	10.55	15500	2.2259	1.2545
0.5083	10.89	16000	2.1764	1.2214
0.3901	11.23	16500	2.2412	1.2341
0.4275	11.57	17000	2.3781	1.2290
0.4225	11.91	17500	2.1578	1.2443
0.4106	12.25	18000	2.5651	1.2341
0.3933	12.59	18500	2.1819	1.2265
0.3821	12.93	19000	2.0564	1.1934
0.3584	13.27	19500	2.5475	1.2290
0.3468	13.61	20000	2.5857	1.1781
0.3984	13.96	20500	2.2383	1.2239
0.308	14.3	21000	2.4947	1.2137
0.3356	14.64	21500	2.6563	1.2163
0.3406	14.98	22000	2.3337	1.2061
0.3297	15.32	22500	2.2793	1.1908
0.3028	15.66	23000	2.6462	1.1654
0.3226	16.0	23500	2.3785	1.1705
0.2605	16.34	24000	2.7212	1.1858
0.2669	16.68	24500	3.0365	1.2087
0.2967	17.02	25000	2.4898	1.1934
0.2547	17.36	25500	2.4020	1.1832
0.2779	17.7	26000	2.5558	1.1705
0.2341	18.04	26500	2.9406	1.1934
0.2304	18.38	27000	3.1528	1.1603
0.226	18.72	27500	3.0001	1.2163
0.2319	19.06	28000	3.0117	1.1603
0.1836	19.4	28500	2.8332	1.1858
0.2085	19.74	29000	2.8757	1.1603
0.2383	20.08	29500	3.2235	1.1934
0.2006	20.42	30000	3.0189	1.1603
0.1722	20.76	30500	2.8001	1.1527
0.1955	21.1	31000	3.0401	1.1578
0.1839	21.44	31500	3.2621	1.1578
0.1592	21.78	32000	3.1740	1.1552
0.1835	22.12	32500	3.3974	1.1934
0.197	22.46	33000	2.8283	1.1425
0.1788	22.8	33500	3.1983	1.1705
0.169	23.14	34000	3.1978	1.1425
0.1649	23.49	34500	3.1829	1.1552
0.1431	23.83	35000	3.0528	1.1272
0.1384	24.17	35500	3.3792	1.1196
0.1234	24.51	36000	3.3988	1.1425
0.1552	24.85	36500	3.1008	1.1170
0.124	25.19	37000	2.9486	1.1374
0.1439	25.53	37500	3.1028	1.1323
0.1612	25.87	38000	3.0209	1.1043
0.1456	26.21	38500	2.9466	1.1323
0.1333	26.55	39000	3.1298	1.1221
0.1368	26.89	39500	3.1051	1.1272
0.1263	27.23	40000	3.2888	1.1298
0.1198	27.57	40500	3.0984	1.1298
0.1202	27.91	41000	3.1653	1.1374
0.1252	28.25	41500	3.3016	1.1552
0.1177	28.59	42000	3.2566	1.1349
0.1072	28.93	42500	3.3303	1.1425
0.1497	29.27	43000	3.2549	1.1399
0.1089	29.61	43500	3.3121	1.1374
0.0936	29.95	44000	3.3358	1.1374

Framework versions

• Transformers 4.23.1
• Pytorch 1.12.1+cu113
• Datasets 1.18.3
• Tokenizers 0.13.2