fine-w2v2base-bs16-ep100-lr2e-05-linguistic-rmsnorm-focal_ctc_a0.99_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: 4.2488
• Wer: 0.0990

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
2181.8592	0.94	50	1087.6210	15.9420
1908.6856	1.89	100	809.7703	15.8773
838.4017	2.83	150	112.6467	0.9997
117.7945	3.77	200	85.6792	1.0
109.9946	4.72	250	82.5771	1.0
105.7306	5.66	300	79.6600	1.0
102.0127	6.6	350	77.2287	1.0
97.9428	7.55	400	75.4334	1.0
96.0055	8.49	450	74.6870	1.0
96.9376	9.43	500	74.2493	1.0
95.6634	10.38	550	74.1341	1.0
96.1578	11.32	600	74.9003	1.0
92.5678	12.26	650	75.6603	1.0598
90.5927	13.21	700	73.4555	1.0539
87.8965	14.15	750	72.4102	0.9987
86.8467	15.09	800	69.7737	0.9984
85.3381	16.04	850	67.8433	0.9717
80.3298	16.98	900	52.4081	0.8594
56.9494	17.92	950	25.2678	0.3554
32.292	18.87	1000	14.8634	0.2190
22.3255	19.81	1050	11.2898	0.1823
17.6187	20.75	1100	9.1387	0.1534
15.1531	21.7	1150	7.6636	0.1368
13.1696	22.64	1200	7.0291	0.1434
11.9792	23.58	1250	6.6867	0.1325
11.2404	24.53	1300	6.2948	0.1213
10.6256	25.47	1350	5.7151	0.1180
9.452	26.42	1400	5.4196	0.1175
9.3087	27.36	1450	5.2929	0.1124
8.5149	28.3	1500	5.1394	0.1163
8.3662	29.25	1550	5.1275	0.1213
7.8852	30.19	1600	4.9033	0.1093
7.5135	31.13	1650	4.9572	0.1097
7.5374	32.08	1700	4.7588	0.1016
7.2968	33.02	1750	4.7317	0.1033
7.0861	33.96	1800	4.7916	0.1087
6.6371	34.91	1850	4.7941	0.1132
6.6186	35.85	1900	4.6608	0.1036
6.6288	36.79	1950	4.6790	0.1074
6.2433	37.74	2000	4.7715	0.1121
6.2362	38.68	2050	4.6420	0.1034
5.957	39.62	2100	4.5756	0.1070
5.8034	40.57	2150	4.4112	0.1060
5.4943	41.51	2200	4.5632	0.1034
5.5593	42.45	2250	4.5376	0.1105
5.3447	43.4	2300	4.5423	0.1006
5.4181	44.34	2350	4.3789	0.0993
5.222	45.28	2400	4.3695	0.1031
5.1146	46.23	2450	4.4108	0.1084
5.0952	47.17	2500	4.2957	0.1016
4.9023	48.11	2550	4.3769	0.1021
5.1633	49.06	2600	4.3633	0.1063
4.9489	50.0	2650	4.3422	0.1045
4.7391	50.94	2700	4.2510	0.1029
4.7996	51.89	2750	4.3254	0.1012
4.244	52.83	2800	4.4121	0.1035
4.5831	53.77	2850	4.4056	0.1044
4.5198	54.72	2900	4.3638	0.1050
4.1964	55.66	2950	4.3397	0.1071
4.0544	56.6	3000	4.3493	0.1031
4.3568	57.55	3050	4.4721	0.1059
4.2692	58.49	3100	4.4278	0.1117
4.1226	59.43	3150	4.3081	0.1004
4.2681	60.38	3200	4.4176	0.1059
3.8412	61.32	3250	4.3213	0.1028
4.1387	62.26	3300	4.3419	0.1056
3.6847	63.21	3350	4.2498	0.1065
3.8768	64.15	3400	4.2776	0.1028
3.659	65.09	3450	4.2988	0.1008
3.809	66.04	3500	4.3041	0.1034
3.7459	66.98	3550	4.2955	0.0995
3.7996	67.92	3600	4.2843	0.0993
3.6773	68.87	3650	4.2396	0.0988
3.6364	69.81	3700	4.2206	0.0963
3.6342	70.75	3750	4.2905	0.1018
3.7012	71.7	3800	4.3084	0.0994
3.4846	72.64	3850	4.2872	0.0976
3.4814	73.58	3900	4.2596	0.1003
3.3212	74.53	3950	4.2270	0.0964
3.6578	75.47	4000	4.2477	0.0978
3.4573	76.42	4050	4.2389	0.0973
3.5776	77.36	4100	4.2827	0.0989
3.5116	78.3	4150	4.3245	0.1002
3.3334	79.25	4200	4.2707	0.0996
3.4829	80.19	4250	4.2456	0.0982
3.44	81.13	4300	4.2846	0.1003
3.4112	82.08	4350	4.2800	0.0977
3.3825	83.02	4400	4.2569	0.0976
3.3444	83.96	4450	4.2334	0.0949
3.5125	84.91	4500	4.2632	0.0978
3.3393	85.85	4550	4.2508	0.0979
3.4698	86.79	4600	4.2483	0.1000
3.3466	87.74	4650	4.2560	0.0985
3.3808	88.68	4700	4.2550	0.0973
3.3442	89.62	4750	4.2574	0.0982
3.0359	90.57	4800	4.2572	0.0993
3.5286	91.51	4850	4.2509	0.0993
3.0826	92.45	4900	4.2408	0.0977
3.513	93.4	4950	4.2531	0.0990
3.272	94.34	5000	4.2558	0.0995
3.2433	95.28	5050	4.2515	0.0992
3.3373	96.23	5100	4.2524	0.1001
3.2239	97.17	5150	4.2540	0.0995
3.4072	98.11	5200	4.2486	0.0993
3.3015	99.06	5250	4.2497	0.0988
3.329	100.0	5300	4.2488	0.0990

Framework versions

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