fine-w2v2base-bs16-ep100-lr2e-05-linguistic-rmsnorm-focal_ctc_a0.99_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: 4.1495
• Wer: 0.0930

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
2149.8069	0.94	50	1029.5103	12.5140
1384.9331	1.89	100	295.9721	0.9978
219.794	2.83	150	86.8886	1.0
113.249	3.77	200	83.8456	1.0
109.1227	4.72	250	81.2277	1.0
105.1573	5.66	300	78.3234	1.0
101.7412	6.6	350	76.3514	1.0
97.6664	7.55	400	74.8664	1.0
95.8132	8.49	450	74.1711	1.0
96.7632	9.43	500	73.7442	1.0
95.3477	10.38	550	73.6445	1.0
95.4528	11.32	600	73.7788	0.9991
91.1317	12.26	650	66.9474	0.9809
71.8284	13.21	700	35.2335	0.4713
40.6304	14.15	750	19.0379	0.2671
26.5956	15.09	800	13.2650	0.2020
20.6269	16.04	850	10.4302	0.1667
17.2297	16.98	900	9.0816	0.1531
14.7348	17.92	950	7.7998	0.1358
13.4356	18.87	1000	7.3014	0.1381
12.2847	19.81	1050	6.9627	0.1386
11.5782	20.75	1100	6.3901	0.1300
11.1732	21.7	1150	6.0007	0.1185
10.2335	22.64	1200	5.9507	0.1261
9.7343	23.58	1250	5.6958	0.1177
9.0428	24.53	1300	5.6682	0.1160
9.117	25.47	1350	5.4908	0.1161
8.4094	26.42	1400	5.3418	0.1135
8.2214	27.36	1450	5.1586	0.1094
7.885	28.3	1500	4.9319	0.1086
7.7676	29.25	1550	5.0031	0.1129
7.4375	30.19	1600	4.9441	0.1100
7.0199	31.13	1650	4.7904	0.1041
7.0727	32.08	1700	4.7495	0.1031
6.6648	33.02	1750	4.6025	0.1018
6.5168	33.96	1800	4.7012	0.1019
6.2194	34.91	1850	4.6766	0.1087
6.15	35.85	1900	4.5767	0.1031
6.1484	36.79	1950	4.4289	0.1064
5.7505	37.74	2000	4.4011	0.0991
5.8478	38.68	2050	4.4077	0.0952
5.5878	39.62	2100	4.4689	0.0989
5.6626	40.57	2150	4.4692	0.0950
5.3951	41.51	2200	4.4790	0.0967
5.3447	42.45	2250	4.3929	0.0974
5.1027	43.4	2300	4.3692	0.0949
5.1015	44.34	2350	4.3436	0.0935
5.0664	45.28	2400	4.2644	0.0956
4.7384	46.23	2450	4.2963	0.0999
4.6469	47.17	2500	4.2131	0.0933
4.5561	48.11	2550	4.2021	0.0952
4.7177	49.06	2600	4.2031	0.0983
4.4587	50.0	2650	4.2315	0.0991
4.3943	50.94	2700	4.2598	0.0953
4.5284	51.89	2750	4.1909	0.0944
4.0457	52.83	2800	4.2877	0.0963
4.2793	53.77	2850	4.2052	0.0953
4.387	54.72	2900	4.2593	0.1024
3.9789	55.66	2950	4.2190	0.0950
3.8419	56.6	3000	4.2314	0.0930
4.0432	57.55	3050	4.2830	0.0983
4.0056	58.49	3100	4.2671	0.1029
3.8839	59.43	3150	4.2807	0.0951
3.9377	60.38	3200	4.3071	0.1009
3.6095	61.32	3250	4.2250	0.0938
3.944	62.26	3300	4.2492	0.1008
3.5562	63.21	3350	4.2156	0.1013
3.6647	64.15	3400	4.2157	0.0974
3.5694	65.09	3450	4.2178	0.0970
3.6198	66.04	3500	4.1781	0.0961
3.5949	66.98	3550	4.1398	0.0929
3.605	67.92	3600	4.1940	0.0969
3.4902	68.87	3650	4.1712	0.0918
3.4942	69.81	3700	4.1447	0.0898
3.4367	70.75	3750	4.1606	0.0944
3.4854	71.7	3800	4.1472	0.0932
3.3036	72.64	3850	4.1874	0.0923
3.2617	73.58	3900	4.1866	0.0941
3.1137	74.53	3950	4.1552	0.0906
3.4462	75.47	4000	4.1435	0.0905
3.2211	76.42	4050	4.1213	0.0935
3.3305	77.36	4100	4.1661	0.0933
3.2492	78.3	4150	4.1404	0.0923
3.0898	79.25	4200	4.1700	0.0928
3.2347	80.19	4250	4.1557	0.0903
3.2544	81.13	4300	4.1916	0.0961
3.1672	82.08	4350	4.1605	0.0918
3.1577	83.02	4400	4.1670	0.0921
3.0994	83.96	4450	4.1541	0.0916
3.2358	84.91	4500	4.1625	0.0917
3.0938	85.85	4550	4.1797	0.0923
3.1622	86.79	4600	4.1639	0.0909
3.2359	87.74	4650	4.1759	0.0938
3.188	88.68	4700	4.1590	0.0913
3.177	89.62	4750	4.1573	0.0912
2.9153	90.57	4800	4.1643	0.0926
3.3507	91.51	4850	4.1631	0.0930
2.8699	92.45	4900	4.1474	0.0913
3.3063	93.4	4950	4.1534	0.0926
3.0762	94.34	5000	4.1586	0.0926
2.9829	95.28	5050	4.1550	0.0928
3.172	96.23	5100	4.1527	0.0930
3.0076	97.17	5150	4.1520	0.0931
3.125	98.11	5200	4.1517	0.0926
3.0391	99.06	5250	4.1495	0.0928
3.2004	100.0	5300	4.1495	0.0930

Framework versions

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