wav2vec2-large-xlsr-Mongolian-cv17-base

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.1550
• Wer: 0.6570

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: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 80
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.5882	20	12.9741	1.0
No log	1.1765	40	12.7936	1.0002
No log	1.7647	60	12.3703	1.0
No log	2.3529	80	5.6880	1.0
No log	2.9412	100	3.6853	1.0
No log	3.5294	120	3.3076	1.0
No log	4.1176	140	3.2023	1.0
No log	4.7059	160	3.1422	1.0
No log	5.2941	180	3.1331	1.0
No log	5.8824	200	3.1183	1.0
No log	6.4706	220	3.1175	1.0
No log	7.0588	240	3.1132	1.0
No log	7.6471	260	3.1111	1.0
No log	8.2353	280	3.1101	1.0
No log	8.8235	300	3.1135	1.0
No log	9.4118	320	3.1039	1.0
No log	10.0	340	3.0961	1.0
No log	10.5882	360	3.0809	1.0
No log	11.1765	380	3.0651	1.0
4.9312	11.7647	400	3.0478	1.0
4.9312	12.3529	420	3.0584	1.0
4.9312	12.9412	440	3.0064	1.0
4.9312	13.5294	460	2.8224	1.0
4.9312	14.1176	480	2.5811	1.0
4.9312	14.7059	500	2.1769	1.0032
4.9312	15.2941	520	1.7646	1.0742
4.9312	15.8824	540	1.4124	1.0159
4.9312	16.4706	560	1.2848	0.9538
4.9312	17.0588	580	1.2267	0.9808
4.9312	17.6471	600	1.1108	0.9423
4.9312	18.2353	620	1.1815	0.9678
4.9312	18.8235	640	1.0553	0.8896
4.9312	19.4118	660	1.0977	0.8884
4.9312	20.0	680	0.9775	0.8532
4.9312	20.5882	700	0.9972	0.8340
4.9312	21.1765	720	1.0438	0.8009
4.9312	21.7647	740	0.9990	0.7850
4.9312	22.3529	760	0.9693	0.7595
4.9312	22.9412	780	1.0659	0.7699
1.3568	23.5294	800	0.9913	0.7610
1.3568	24.1176	820	1.0340	0.7547
1.3568	24.7059	840	1.0347	0.7337
1.3568	25.2941	860	1.0703	0.7437
1.3568	25.8824	880	1.0441	0.7350
1.3568	26.4706	900	1.0683	0.7261
1.3568	27.0588	920	1.0231	0.7296
1.3568	27.6471	940	1.0517	0.7291
1.3568	28.2353	960	1.1089	0.7417
1.3568	28.8235	980	1.0957	0.7223
1.3568	29.4118	1000	1.1120	0.7258
1.3568	30.0	1020	1.0992	0.7396
1.3568	30.5882	1040	1.1502	0.7190
1.3568	31.1765	1060	1.0743	0.7225
1.3568	31.7647	1080	1.0548	0.7178
1.3568	32.3529	1100	1.0534	0.7104
1.3568	32.9412	1120	1.0752	0.7083
1.3568	33.5294	1140	1.1574	0.7160
1.3568	34.1176	1160	1.1471	0.7190
1.3568	34.7059	1180	1.1077	0.7093
0.2559	35.2941	1200	1.0737	0.7004
0.2559	35.8824	1220	1.0822	0.6905
0.2559	36.4706	1240	1.0836	0.6889
0.2559	37.0588	1260	1.1399	0.6975
0.2559	37.6471	1280	1.0981	0.6880
0.2559	38.2353	1300	1.0887	0.6938
0.2559	38.8235	1320	1.0870	0.7112
0.2559	39.4118	1340	1.1324	0.6978
0.2559	40.0	1360	1.1170	0.6834
0.2559	40.5882	1380	1.1032	0.6761
0.2559	41.1765	1400	1.1361	0.7035
0.2559	41.7647	1420	1.0855	0.6965
0.2559	42.3529	1440	1.1320	0.6933
0.2559	42.9412	1460	1.1194	0.6849
0.2559	43.5294	1480	1.0870	0.6912
0.2559	44.1176	1500	1.1434	0.6785
0.2559	44.7059	1520	1.1434	0.6926
0.2559	45.2941	1540	1.1703	0.6839
0.2559	45.8824	1560	1.1275	0.6762
0.2559	46.4706	1580	1.1511	0.6840
0.1626	47.0588	1600	1.1336	0.6771
0.1626	47.6471	1620	1.1421	0.6785
0.1626	48.2353	1640	1.1084	0.6831
0.1626	48.8235	1660	1.1682	0.6831
0.1626	49.4118	1680	1.1349	0.6763
0.1626	50.0	1700	1.1561	0.6793
0.1626	50.5882	1720	1.1117	0.6660
0.1626	51.1765	1740	1.1875	0.6834
0.1626	51.7647	1760	1.1453	0.6782
0.1626	52.3529	1780	1.1040	0.6744
0.1626	52.9412	1800	1.1213	0.6711
0.1626	53.5294	1820	1.1454	0.6689
0.1626	54.1176	1840	1.1659	0.6706
0.1626	54.7059	1860	1.1616	0.6823
0.1626	55.2941	1880	1.2440	0.6817
0.1626	55.8824	1900	1.1472	0.6753
0.1626	56.4706	1920	1.1588	0.6691
0.1626	57.0588	1940	1.1590	0.6731
0.1626	57.6471	1960	1.1649	0.6712
0.1626	58.2353	1980	1.1990	0.6680
0.123	58.8235	2000	1.1282	0.6681
0.123	59.4118	2020	1.1609	0.6686
0.123	60.0	2040	1.1722	0.6703
0.123	60.5882	2060	1.1538	0.6739
0.123	61.1765	2080	1.1679	0.6727
0.123	61.7647	2100	1.1747	0.6687
0.123	62.3529	2120	1.1716	0.6691
0.123	62.9412	2140	1.1785	0.6655
0.123	63.5294	2160	1.1485	0.6658
0.123	64.1176	2180	1.1578	0.6626
0.123	64.7059	2200	1.1694	0.6648
0.123	65.2941	2220	1.1711	0.6677
0.123	65.8824	2240	1.1581	0.6624
0.123	66.4706	2260	1.1650	0.6723
0.123	67.0588	2280	1.1789	0.6637
0.123	67.6471	2300	1.1705	0.6624
0.123	68.2353	2320	1.1071	0.6615
0.123	68.8235	2340	1.1300	0.6654
0.123	69.4118	2360	1.1616	0.6672
0.123	70.0	2380	1.1671	0.6568
0.0991	70.5882	2400	1.1493	0.6587
0.0991	71.1765	2420	1.1476	0.6575
0.0991	71.7647	2440	1.1691	0.6582
0.0991	72.3529	2460	1.1867	0.6609
0.0991	72.9412	2480	1.1427	0.6519
0.0991	73.5294	2500	1.1635	0.6558
0.0991	74.1176	2520	1.1503	0.6553
0.0991	74.7059	2540	1.1487	0.6562
0.0991	75.2941	2560	1.1552	0.6576
0.0991	75.8824	2580	1.1638	0.6586
0.0991	76.4706	2600	1.1601	0.6566
0.0991	77.0588	2620	1.1603	0.6558
0.0991	77.6471	2640	1.1564	0.6547
0.0991	78.2353	2660	1.1560	0.6556
0.0991	78.8235	2680	1.1550	0.6564
0.0991	79.4118	2700	1.1550	0.6567
0.0991	80.0	2720	1.1550	0.6570

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1