fine-w2v2base-bs16-ep100-lr2e-05-linguistic-rmsnorm-focal_ctc_a0.75_g1.0-0.05_10_0.004_40

This model is a fine-tuned version of nguyenvulebinh/wav2vec2-base-vietnamese-250h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.1907
• Wer: 0.1001

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 64
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1628.3266	0.94	50	795.4863	15.7055
1167.4639	1.89	100	276.5289	0.9954
193.5269	2.83	150	67.2639	1.0
86.2802	3.77	200	63.6259	1.0
82.8944	4.72	250	61.8939	1.0
79.9575	5.66	300	59.7914	1.0
77.3429	6.6	350	58.1507	1.0
74.0584	7.55	400	57.2977	1.0
72.6694	8.49	450	56.7115	1.0
73.417	9.43	500	56.6074	1.0
72.4291	10.38	550	56.4755	1.0
72.7847	11.32	600	56.8623	1.0
69.3297	12.26	650	49.6540	0.9637
54.2644	13.21	700	29.1559	0.5631
31.2303	14.15	750	13.8957	0.2414
19.4522	15.09	800	9.7460	0.1949
15.0046	16.04	850	7.6735	0.1622
12.3783	16.98	900	6.5559	0.1520
10.7256	17.92	950	5.7852	0.1423
9.8218	18.87	1000	5.4473	0.1395
9.0115	19.81	1050	5.1250	0.1356
8.1076	20.75	1100	4.7980	0.1233
7.9779	21.7	1150	4.6150	0.1211
7.6027	22.64	1200	4.6507	0.1251
7.4535	23.58	1250	4.4814	0.1210
6.946	24.53	1300	4.4369	0.1149
7.0627	25.47	1350	4.1153	0.1139
6.2482	26.42	1400	4.0045	0.1101
6.2238	27.36	1450	4.0355	0.1158
5.8919	28.3	1500	3.9625	0.1154
5.7955	29.25	1550	3.7957	0.1127
5.4849	30.19	1600	3.7986	0.1058
5.1108	31.13	1650	3.8188	0.1070
5.3354	32.08	1700	3.6909	0.1024
5.1149	33.02	1750	3.6227	0.1023
4.976	33.96	1800	3.6176	0.1016
4.5904	34.91	1850	3.5959	0.1079
4.6613	35.85	1900	3.5000	0.1069
4.7697	36.79	1950	3.5211	0.1014
4.4224	37.74	2000	3.4720	0.1001
4.5255	38.68	2050	3.4178	0.0983
4.2808	39.62	2100	3.4801	0.1044
4.2407	40.57	2150	3.4080	0.1000
3.9611	41.51	2200	3.4514	0.1049
4.014	42.45	2250	3.3983	0.1089
3.8487	43.4	2300	3.4164	0.1042
3.8132	44.34	2350	3.3562	0.0958
3.6973	45.28	2400	3.2839	0.0978
3.606	46.23	2450	3.3125	0.1009
3.5412	47.17	2500	3.2580	0.0977
3.3971	48.11	2550	3.3065	0.0984
3.4795	49.06	2600	3.3312	0.1037
3.302	50.0	2650	3.3015	0.0986
3.2486	50.94	2700	3.2506	0.0977
3.3977	51.89	2750	3.2406	0.0952
3.0229	52.83	2800	3.2880	0.0989
3.2615	53.77	2850	3.3112	0.0998
3.2023	54.72	2900	3.2895	0.1037
3.0037	55.66	2950	3.3394	0.1018
2.9249	56.6	3000	3.2351	0.0974
3.112	57.55	3050	3.2868	0.1019
3.0261	58.49	3100	3.3241	0.1039
2.8959	59.43	3150	3.2251	0.0947
2.946	60.38	3200	3.2880	0.1012
2.6933	61.32	3250	3.2595	0.1031
2.8755	62.26	3300	3.2140	0.1048
2.606	63.21	3350	3.2743	0.1075
2.7607	64.15	3400	3.2455	0.1053
2.6394	65.09	3450	3.2335	0.0994
2.6899	66.04	3500	3.2278	0.1004
2.719	66.98	3550	3.2012	0.0979
2.6997	67.92	3600	3.2009	0.0979
2.5935	68.87	3650	3.2141	0.0978
2.6115	69.81	3700	3.1760	0.0947
2.5713	70.75	3750	3.1937	0.0977
2.6647	71.7	3800	3.1629	0.0986
2.4878	72.64	3850	3.1675	0.0952
2.4761	73.58	3900	3.1951	0.0976
2.3124	74.53	3950	3.1629	0.0954
2.5718	75.47	4000	3.1577	0.0978
2.4606	76.42	4050	3.1632	0.0973
2.5313	77.36	4100	3.1841	0.0988
2.5124	78.3	4150	3.1894	0.0987
2.3324	79.25	4200	3.1719	0.0966
2.4468	80.19	4250	3.1760	0.0964
2.4035	81.13	4300	3.2014	0.0983
2.3834	82.08	4350	3.1823	0.0966
2.3655	83.02	4400	3.1758	0.0948
2.3525	83.96	4450	3.1921	0.0980
2.4428	84.91	4500	3.1990	0.0970
2.3276	85.85	4550	3.1907	0.0984
2.4423	86.79	4600	3.1893	0.0977
2.3457	87.74	4650	3.2001	0.1005
2.4146	88.68	4700	3.1883	0.0985
2.3415	89.62	4750	3.1934	0.0997
2.2057	90.57	4800	3.1939	0.0995
2.5141	91.51	4850	3.1944	0.1006
2.175	92.45	4900	3.1808	0.0986
2.4668	93.4	4950	3.1885	0.0994
2.2732	94.34	5000	3.1877	0.0998
2.2636	95.28	5050	3.1877	0.0989
2.3504	96.23	5100	3.1904	0.1000
2.2721	97.17	5150	3.1917	0.1005
2.4014	98.11	5200	3.1922	0.1003
2.3263	99.06	5250	3.1897	0.0998
2.3731	100.0	5300	3.1907	0.1001

Framework versions

• Transformers 4.34.0
• Pytorch 2.0.1
• Datasets 2.14.5
• Tokenizers 0.14.1