fine-w2v2base-bs16-ep100-lr2e-05-linguistic-rmsnorm-focal_ctc_a0.75_g0.5-0.05_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: 3.1140
• Wer: 0.0915

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
1629.9275	0.94	50	824.7193	15.8408
1138.8538	1.89	100	363.0830	1.0609
228.953	2.83	150	77.0798	1.0
87.3839	3.77	200	66.6220	1.0
83.3628	4.72	250	64.3104	1.0
79.8819	5.66	300	61.3996	1.0
76.9216	6.6	350	58.9846	1.0
73.8162	7.55	400	57.3442	1.0
72.5154	8.49	450	56.7676	1.0
73.3129	9.43	500	56.7173	1.0
72.2926	10.38	550	56.5543	1.0
72.4577	11.32	600	58.1313	1.0
69.5175	12.26	650	57.4489	1.0076
64.6635	13.21	700	41.0107	0.7747
42.1225	14.15	750	18.6933	0.3211
23.9017	15.09	800	11.5678	0.2335
17.2962	16.04	850	8.6803	0.1841
13.8834	16.98	900	7.2569	0.1655
11.6255	17.92	950	6.2023	0.1497
10.4288	18.87	1000	5.5896	0.1394
9.5611	19.81	1050	5.3111	0.1419
8.7185	20.75	1100	5.0459	0.1333
8.529	21.7	1150	4.6049	0.1241
7.9187	22.64	1200	4.4407	0.1241
7.3237	23.58	1250	4.2262	0.1135
6.9945	24.53	1300	4.2348	0.1133
6.9508	25.47	1350	3.9280	0.1054
6.3118	26.42	1400	3.8789	0.1085
6.3038	27.36	1450	3.9444	0.1125
5.9028	28.3	1500	3.8333	0.1078
5.9109	29.25	1550	3.8047	0.1060
5.8046	30.19	1600	3.7575	0.1144
5.5068	31.13	1650	3.6156	0.0993
5.4652	32.08	1700	3.6463	0.1033
5.1792	33.02	1750	3.5317	0.1018
5.2711	33.96	1800	3.5806	0.1042
4.764	34.91	1850	3.5744	0.1024
4.8339	35.85	1900	3.4476	0.0966
4.7665	36.79	1950	3.3453	0.0989
4.4695	37.74	2000	3.3646	0.0933
4.5748	38.68	2050	3.4034	0.1019
4.3533	39.62	2100	3.4187	0.1035
4.2584	40.57	2150	3.3029	0.0993
4.0446	41.51	2200	3.3336	0.0972
4.1068	42.45	2250	3.3550	0.0993
3.9195	43.4	2300	3.3538	0.0998
3.9058	44.34	2350	3.2872	0.0960
3.8691	45.28	2400	3.3699	0.1010
3.6487	46.23	2450	3.3958	0.1033
3.7089	47.17	2500	3.4632	0.1034
3.5368	48.11	2550	3.2808	0.0961
3.6149	49.06	2600	3.3465	0.1019
3.4101	50.0	2650	3.2952	0.0970
3.392	50.94	2700	3.1991	0.0947
3.5055	51.89	2750	3.2169	0.0958
3.0548	52.83	2800	3.2389	0.1014
3.3108	53.77	2850	3.2238	0.0963
3.2846	54.72	2900	3.2196	0.1016
3.0562	55.66	2950	3.2425	0.1014
2.9703	56.6	3000	3.1926	0.0960
3.15	57.55	3050	3.2608	0.1019
3.1351	58.49	3100	3.2207	0.0999
3.0213	59.43	3150	3.1639	0.0973
3.0526	60.38	3200	3.2448	0.1008
2.7631	61.32	3250	3.1578	0.0909
2.9872	62.26	3300	3.1629	0.0953
2.7601	63.21	3350	3.1266	0.0967
2.8478	64.15	3400	3.1390	0.0939
2.726	65.09	3450	3.1591	0.0961
2.7968	66.04	3500	3.1354	0.0961
2.7528	66.98	3550	3.1616	0.0973
2.7885	67.92	3600	3.1367	0.0913
2.6265	68.87	3650	3.1837	0.0948
2.6711	69.81	3700	3.1300	0.0911
2.6724	70.75	3750	3.1289	0.0943
2.7063	71.7	3800	3.1347	0.0958
2.52	72.64	3850	3.1297	0.0934
2.5192	73.58	3900	3.1147	0.0918
2.385	74.53	3950	3.1021	0.0913
2.6387	75.47	4000	3.1284	0.0918
2.534	76.42	4050	3.1065	0.0919
2.5553	77.36	4100	3.1210	0.0953
2.5418	78.3	4150	3.1205	0.0928
2.3757	79.25	4200	3.1181	0.0926
2.5093	80.19	4250	3.0970	0.0922
2.4721	81.13	4300	3.1469	0.0938
2.4406	82.08	4350	3.1273	0.0918
2.4254	83.02	4400	3.1289	0.0907
2.4009	83.96	4450	3.1118	0.0897
2.5242	84.91	4500	3.0989	0.0911
2.4325	85.85	4550	3.1187	0.0922
2.5331	86.79	4600	3.0940	0.0921
2.4234	87.74	4650	3.0955	0.0917
2.4607	88.68	4700	3.1024	0.0925
2.407	89.62	4750	3.1032	0.0923
2.2203	90.57	4800	3.1189	0.0912
2.5802	91.51	4850	3.1072	0.0917
2.2169	92.45	4900	3.1065	0.0908
2.5712	93.4	4950	3.1111	0.0914
2.393	94.34	5000	3.1136	0.0916
2.3262	95.28	5050	3.1137	0.0918
2.4033	96.23	5100	3.1175	0.0911
2.3637	97.17	5150	3.1156	0.0915
2.4371	98.11	5200	3.1153	0.0915
2.4	99.06	5250	3.1138	0.0914
2.4233	100.0	5300	3.1140	0.0915

Framework versions

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