exp17-F03-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: 1.9268
• Wer: 0.9485

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
47.4704	0.36	500	3.3075	1.0131
3.1649	0.71	1000	3.3442	1.0
2.9674	1.07	1500	2.6986	1.0
2.7514	1.42	2000	2.5789	1.1299
2.6045	1.78	2500	2.3025	1.2529
2.373	2.14	3000	2.2169	1.2698
2.1632	2.49	3500	1.9883	1.2667
2.0942	2.85	4000	1.9294	1.2567
1.9239	3.2	4500	1.9799	1.2467
1.7549	3.56	5000	1.7485	1.2252
1.6973	3.91	5500	1.6799	1.2283
1.5823	4.27	6000	1.6847	1.2267
1.4761	4.63	6500	1.6971	1.1968
1.4381	4.98	7000	1.6280	1.2052
1.2509	5.34	7500	1.6657	1.2060
1.3112	5.69	8000	1.5618	1.1783
1.1851	6.05	8500	1.6555	1.1783
1.1112	6.41	9000	1.6586	1.1752
1.0463	6.76	9500	1.6135	1.1683
1.041	7.12	10000	1.5444	1.1522
0.9451	7.47	10500	1.5561	1.1622
0.9454	7.83	11000	1.5044	1.1483
0.8496	8.19	11500	1.6724	1.1330
0.825	8.54	12000	1.5950	1.1414
0.8291	8.9	12500	1.6023	1.1384
0.7279	9.25	13000	1.6319	1.1314
0.7394	9.61	13500	1.5478	1.1337
0.7079	9.96	14000	1.7564	1.1453
0.609	10.32	14500	1.7671	1.1245
0.6639	10.68	15000	1.7471	1.1314
0.648	11.03	15500	1.7694	1.2160
0.577	11.39	16000	1.6149	1.1760
0.577	11.74	16500	1.9288	1.1238
0.5695	12.1	17000	1.7503	1.1253
0.5326	12.46	17500	1.5635	1.1376
0.5423	12.81	18000	1.7083	1.1668
0.4775	13.17	18500	1.7054	1.1245
0.4772	13.52	19000	1.6455	1.1045
0.4737	13.88	19500	1.5996	1.0968
0.4529	14.23	20000	1.9847	1.1653
0.4461	14.59	20500	1.6845	1.1084
0.4497	14.95	21000	1.6465	1.0938
0.4096	15.3	21500	1.5919	1.0769
0.3897	15.66	22000	1.5637	1.0761
0.4234	16.01	22500	1.6360	1.0953
0.3659	16.37	23000	1.7573	1.0830
0.3352	16.73	23500	1.8474	1.0976
0.3886	17.08	24000	1.9115	1.0953
0.3255	17.44	24500	1.8820	1.0815
0.3405	17.79	25000	1.6862	1.0346
0.3205	18.15	25500	1.6912	1.0500
0.322	18.51	26000	1.6253	1.0615
0.296	18.86	26500	1.7924	1.0546
0.2869	19.22	27000	1.8204	1.0899
0.269	19.57	27500	1.7558	1.0292
0.2844	19.93	28000	1.6038	1.0131
0.2543	20.28	28500	1.7935	1.0161
0.3025	20.64	29000	1.8706	1.0423
0.2707	21.0	29500	2.0011	1.0208
0.2401	21.35	30000	1.9058	1.0161
0.2609	21.71	30500	1.7555	1.0015
0.2403	22.06	31000	1.9301	1.0085
0.2538	22.42	31500	1.8586	0.9969
0.2334	22.78	32000	1.8588	0.9985
0.2013	23.13	32500	1.9307	1.0108
0.2122	23.49	33000	1.8830	0.9908
0.2242	23.84	33500	1.8133	0.9754
0.188	24.2	34000	1.8435	0.9800
0.2142	24.56	34500	1.8491	0.9792
0.2059	24.91	35000	1.8005	0.9754
0.1794	25.27	35500	1.8845	0.9700
0.185	25.62	36000	1.8620	0.9731
0.1843	25.98	36500	1.8461	0.9539
0.1717	26.33	37000	1.8100	0.9639
0.164	26.69	37500	1.8192	0.9547
0.1888	27.05	38000	1.8005	0.9470
0.1792	27.4	38500	1.8901	0.9562
0.1708	27.76	39000	1.8306	0.9547
0.1508	28.11	39500	1.8934	0.9508
0.1751	28.47	40000	1.8956	0.9523
0.1541	28.83	40500	1.9360	0.9416
0.1611	29.18	41000	1.9346	0.9454
0.1684	29.54	41500	1.9247	0.9470
0.1463	29.89	42000	1.9268	0.9485

Framework versions

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