thunninoi/wav2vec2-japanese-hiragana-vtuber

checkpoints

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

• Loss: 0.4134
• Wer: 0.1884

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: 0.0003
• train_batch_size: 3
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 6
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 75
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4299	1.0	247	0.7608	0.4853
0.8045	2.0	494	0.6603	0.4449
0.6061	3.0	741	0.5527	0.4233
0.4372	4.0	988	0.6262	0.4029
0.3226	5.0	1235	0.4528	0.3462
0.2581	6.0	1482	0.4961	0.3226
0.2147	7.0	1729	0.4856	0.3075
0.1736	8.0	1976	0.4372	0.3063
0.1488	9.0	2223	0.3771	0.2761
0.1286	10.0	2470	0.4373	0.2590
0.1118	11.0	2717	0.3840	0.2594
0.1037	12.0	2964	0.4241	0.2590
0.0888	13.0	3211	0.4150	0.2410
0.0923	14.0	3458	0.3811	0.2524
0.0813	15.0	3705	0.4164	0.2459
0.0671	16.0	3952	0.3498	0.2288
0.0669	17.0	4199	0.3697	0.2247
0.0586	18.0	4446	0.3550	0.2251
0.0533	19.0	4693	0.4024	0.2231
0.0542	20.0	4940	0.4130	0.2121
0.0532	21.0	5187	0.3464	0.2231
0.0451	22.0	5434	0.3346	0.1966
0.0413	23.0	5681	0.4599	0.2088
0.0401	24.0	5928	0.4031	0.2162
0.0345	25.0	6175	0.3726	0.2084
0.033	26.0	6422	0.4619	0.2076
0.0366	27.0	6669	0.4071	0.2202
0.0343	28.0	6916	0.4114	0.2088
0.0319	29.0	7163	0.3605	0.2015
0.0304	30.0	7410	0.4097	0.2015
0.0253	31.0	7657	0.4152	0.1970
0.0235	32.0	7904	0.3829	0.2043
0.0255	33.0	8151	0.3976	0.2011
0.0201	34.0	8398	0.4247	0.2088
0.022	35.0	8645	0.3831	0.1945
0.0175	36.0	8892	0.3838	0.2007
0.0201	37.0	9139	0.4377	0.1986
0.0176	38.0	9386	0.4546	0.2043
0.021	39.0	9633	0.4341	0.2039
0.0191	40.0	9880	0.4043	0.1937
0.0159	41.0	10127	0.4098	0.2064
0.0148	42.0	10374	0.4027	0.1905
0.0129	43.0	10621	0.4104	0.1933
0.0123	44.0	10868	0.3738	0.1925
0.0159	45.0	11115	0.3946	0.1933
0.0091	46.0	11362	0.3971	0.1880
0.0082	47.0	11609	0.4042	0.1986
0.0108	48.0	11856	0.4092	0.1884
0.0123	49.0	12103	0.3674	0.1941
0.01	50.0	12350	0.3750	0.1876
0.0094	51.0	12597	0.3781	0.1831
0.008	52.0	12844	0.4051	0.1852
0.0079	53.0	13091	0.3981	0.1937
0.0068	54.0	13338	0.4425	0.1929
0.0061	55.0	13585	0.4183	0.1986
0.0074	56.0	13832	0.3502	0.1880
0.0071	57.0	14079	0.3908	0.1892
0.0079	58.0	14326	0.3908	0.1913
0.0042	59.0	14573	0.3801	0.1864
0.0049	60.0	14820	0.4065	0.1839
0.0063	61.0	15067	0.4170	0.1900
0.0049	62.0	15314	0.3903	0.1856
0.0031	63.0	15561	0.4042	0.1896
0.0054	64.0	15808	0.3890	0.1839
0.0061	65.0	16055	0.3831	0.1847
0.0052	66.0	16302	0.3898	0.1847
0.0032	67.0	16549	0.4230	0.1831
0.0017	68.0	16796	0.4241	0.1823
0.0022	69.0	17043	0.4360	0.1856
0.0026	70.0	17290	0.4233	0.1815
0.0028	71.0	17537	0.4225	0.1835
0.0018	72.0	17784	0.4163	0.1856
0.0034	73.0	18031	0.4120	0.1876
0.0019	74.0	18278	0.4129	0.1876
0.0023	75.0	18525	0.4134	0.1884

Framework versions

• Transformers 4.17.0
• Pytorch 1.11.0+cu113
• Datasets 1.18.3
• Tokenizers 0.12.1