wav2vec2-commonvoice-20subset-xlsr-53-gpu2

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

• Loss: 1.7751
• Wer: 0.9378
• Cer: 0.2802

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: 5e-05
• train_batch_size: 13
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 26
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 300

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	1.9	400	32.9239	1.0	1.0
69.7146	3.81	800	6.6878	1.0	1.0
7.0732	5.71	1200	6.4976	1.0	1.0
6.4558	7.62	1600	6.4214	1.0	1.0
6.2755	9.52	2000	6.2492	1.0143	0.9682
6.2755	11.43	2400	5.8545	1.0525	0.9396
5.8857	13.33	2800	4.4603	1.0742	0.7200
4.54	15.24	3200	3.7454	1.0297	0.6146
3.5614	17.14	3600	3.2387	1.0126	0.5582
2.9773	19.05	4000	2.8934	1.0068	0.5186
2.9773	20.95	4400	2.6116	0.9977	0.4880
2.5488	22.86	4800	2.4307	0.9932	0.4716
2.2665	24.76	5200	2.2844	0.9874	0.4532
2.0508	26.67	5600	2.1050	0.9886	0.4270
1.7944	28.57	6000	1.9768	0.9857	0.4150
1.7944	30.48	6400	1.8712	0.9789	0.3984
1.6074	32.38	6800	1.8050	0.9749	0.3916
1.4656	34.29	7200	1.7572	0.9783	0.3824
1.3429	36.19	7600	1.6546	0.9686	0.3677
1.2215	38.1	8000	1.6265	0.9726	0.3653
1.2215	40.0	8400	1.6046	0.9640	0.3625
1.1133	41.9	8800	1.5787	0.9737	0.3601
1.0702	43.81	9200	1.5449	0.9652	0.3503
0.9732	45.71	9600	1.5307	0.9572	0.3473
0.8858	47.62	10000	1.4962	0.9566	0.3394
0.8858	49.52	10400	1.5053	0.9561	0.3423
0.8067	51.43	10800	1.5053	0.9595	0.3389
0.7418	53.33	11200	1.4833	0.9566	0.3321
0.6962	55.24	11600	1.4927	0.9583	0.3312
0.6395	57.14	12000	1.4833	0.9509	0.3263
0.6395	59.05	12400	1.4908	0.9543	0.3263
0.5834	60.95	12800	1.4937	0.9521	0.3244
0.5422	62.86	13200	1.5123	0.9498	0.3206
0.4907	64.76	13600	1.5149	0.9515	0.3216
0.4525	66.67	14000	1.5079	0.9475	0.3206
0.4525	68.57	14400	1.5305	0.9469	0.3167
0.4229	70.48	14800	1.5427	0.9532	0.3235
0.3835	72.38	15200	1.5402	0.9452	0.3143
0.3642	74.29	15600	1.5569	0.9475	0.3151
0.3378	76.19	16000	1.5744	0.9463	0.3169
0.3378	78.1	16400	1.5578	0.9503	0.3122
0.3238	80.0	16800	1.5748	0.9481	0.3116
0.2997	81.9	17200	1.5708	0.9509	0.3139
0.2841	83.81	17600	1.5944	0.9521	0.3128
0.2573	85.71	18000	1.5941	0.9543	0.3108
0.2573	87.62	18400	1.6095	0.9515	0.3095
0.2496	89.52	18800	1.6170	0.9475	0.3102
0.2342	91.43	19200	1.6399	0.9469	0.3130
0.2261	93.33	19600	1.6241	0.9475	0.3099
0.2062	95.24	20000	1.6309	0.9446	0.3098
0.2062	97.14	20400	1.6360	0.9521	0.3061
0.2009	99.05	20800	1.6280	0.9526	0.3081
0.1916	100.95	21200	1.6606	0.9452	0.3053
0.1841	102.86	21600	1.6677	0.9475	0.3030
0.1794	104.76	22000	1.6625	0.9475	0.3039
0.1794	106.67	22400	1.6524	0.9481	0.3061
0.1718	108.57	22800	1.6761	0.9469	0.3085
0.174	110.48	23200	1.6778	0.9543	0.3048
0.1586	112.38	23600	1.6784	0.9503	0.3024
0.1595	114.29	24000	1.6844	0.9543	0.3021
0.1595	116.19	24400	1.6888	0.9463	0.3035
0.1494	118.1	24800	1.6767	0.9498	0.2984
0.141	120.0	25200	1.6898	0.9441	0.3044
0.139	121.9	25600	1.6812	0.9463	0.2990
0.1361	123.81	26000	1.6965	0.9446	0.2997
0.1361	125.71	26400	1.7046	0.9435	0.3014
0.1285	127.62	26800	1.6941	0.9463	0.2988
0.1273	129.52	27200	1.6980	0.9492	0.3008
0.1215	131.43	27600	1.7161	0.9424	0.2988
0.1188	133.33	28000	1.7033	0.9424	0.2976
0.1188	135.24	28400	1.7159	0.9446	0.2966
0.1183	137.14	28800	1.7157	0.9424	0.2965
0.118	139.05	29200	1.7073	0.9429	0.2932
0.1081	140.95	29600	1.7453	0.9424	0.2979
0.1064	142.86	30000	1.7120	0.9441	0.2964
0.1064	144.76	30400	1.7219	0.9418	0.2970
0.1028	146.67	30800	1.7217	0.9458	0.2960
0.1008	148.57	31200	1.7296	0.9481	0.2965
0.101	150.48	31600	1.7179	0.9412	0.2939
0.096	152.38	32000	1.7267	0.9418	0.2928
0.096	154.29	32400	1.7336	0.9401	0.2938
0.0898	156.19	32800	1.7229	0.9338	0.2921
0.0934	158.1	33200	1.7236	0.9406	0.2907
0.09	160.0	33600	1.7300	0.9378	0.2954
0.09	161.9	34000	1.7358	0.9435	0.2927
0.09	163.81	34400	1.7349	0.9452	0.2948
0.0886	165.71	34800	1.7336	0.9475	0.2935
0.0854	167.62	35200	1.7307	0.9429	0.2906
0.0829	169.52	35600	1.7329	0.9446	0.2947
0.0868	171.43	36000	1.7490	0.9446	0.2905
0.0868	173.33	36400	1.7322	0.9418	0.2929
0.0832	175.24	36800	1.7477	0.9441	0.2924
0.0792	177.14	37200	1.7541	0.9418	0.2897
0.0774	179.05	37600	1.7504	0.9424	0.2908
0.0754	180.95	38000	1.7516	0.9458	0.2925
0.0754	182.86	38400	1.7633	0.9469	0.2912
0.0779	184.76	38800	1.7526	0.9429	0.2928
0.0733	186.67	39200	1.7387	0.9412	0.2916
0.0765	188.57	39600	1.7464	0.9412	0.2900
0.0725	190.48	40000	1.7581	0.9384	0.2887
0.0725	192.38	40400	1.7424	0.9429	0.2872
0.0701	194.29	40800	1.7372	0.9401	0.2887
0.0707	196.19	41200	1.7570	0.9424	0.2904
0.0679	198.1	41600	1.7523	0.9418	0.2896
0.0649	200.0	42000	1.7767	0.9389	0.2891
0.0649	201.9	42400	1.7509	0.9412	0.2875
0.0654	203.81	42800	1.7480	0.9446	0.2878
0.0652	205.71	43200	1.7489	0.9395	0.2866
0.0642	207.62	43600	1.7609	0.9446	0.2871
0.0665	209.52	44000	1.7644	0.9412	0.2887
0.0665	211.43	44400	1.7583	0.9366	0.2882
0.0591	213.33	44800	1.7510	0.9384	0.2869
0.0593	215.24	45200	1.7632	0.9406	0.2874
0.0654	217.14	45600	1.7562	0.9418	0.2864
0.0571	219.05	46000	1.7585	0.9389	0.2850
0.0571	220.95	46400	1.7542	0.9389	0.2853
0.0576	222.86	46800	1.7625	0.9395	0.2857
0.0564	224.76	47200	1.7652	0.9384	0.2857
0.0566	226.67	47600	1.7698	0.9395	0.2829
0.0539	228.57	48000	1.7684	0.9378	0.2845
0.0539	230.48	48400	1.7737	0.9395	0.2849
0.0539	232.38	48800	1.7549	0.9349	0.2832
0.0533	234.29	49200	1.7548	0.9401	0.2839
0.0534	236.19	49600	1.7661	0.9372	0.2841
0.0514	238.1	50000	1.7680	0.9361	0.2844
0.0514	240.0	50400	1.7620	0.9384	0.2854
0.0504	241.9	50800	1.7744	0.9384	0.2841
0.0522	243.81	51200	1.7774	0.9344	0.2838
0.0486	245.71	51600	1.7739	0.9384	0.2839
0.0497	247.62	52000	1.7732	0.9389	0.2840
0.0497	249.52	52400	1.7705	0.9401	0.2842
0.0489	251.43	52800	1.7707	0.9424	0.2841
0.0496	253.33	53200	1.7754	0.9412	0.2830
0.0478	255.24	53600	1.7684	0.9429	0.2830
0.0515	257.14	54000	1.7675	0.9418	0.2815
0.0515	259.05	54400	1.7745	0.9429	0.2819
0.0474	260.95	54800	1.7783	0.9378	0.2820
0.0476	262.86	55200	1.7744	0.9418	0.2813
0.0448	264.76	55600	1.7715	0.9406	0.2822
0.0462	266.67	56000	1.7708	0.9389	0.2822
0.0462	268.57	56400	1.7719	0.9395	0.2820
0.0443	270.48	56800	1.7777	0.9384	0.2816
0.0445	272.38	57200	1.7750	0.9372	0.2807
0.0425	274.29	57600	1.7819	0.9412	0.2820
0.0449	276.19	58000	1.7765	0.9406	0.2807
0.0449	278.1	58400	1.7783	0.9366	0.2803
0.0434	280.0	58800	1.7719	0.9424	0.2809
0.0447	281.9	59200	1.7700	0.9355	0.2802
0.0465	283.81	59600	1.7747	0.9395	0.2802
0.0447	285.71	60000	1.7764	0.9384	0.2811
0.0447	287.62	60400	1.7799	0.9378	0.2807
0.0432	289.52	60800	1.7800	0.9384	0.2809
0.0431	291.43	61200	1.7785	0.9389	0.2807
0.0422	293.33	61600	1.7792	0.9395	0.2811
0.0418	295.24	62000	1.7749	0.9384	0.2807
0.0418	297.14	62400	1.7738	0.9384	0.2805
0.0416	299.05	62800	1.7751	0.9378	0.2802

Framework versions

• Transformers 4.31.0
• Pytorch 1.13.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3