ED_small_cv_v2

This model is a fine-tuned version of on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.0688
• Cer: 0.0677
• Wer: 0.1598
• Mer: 0.1565
• Wil: 0.2593
• Wip: 0.7407
• Hits: 127573
• Substitutions: 17637
• Deletions: 2971
• Insertions: 3069

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: 0.001
• train_batch_size: 128
• eval_batch_size: 64
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 512
• total_eval_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 15000
• num_epochs: 70.0

Training results

Training Loss	Epoch	Step	Cer	Deletions	Hits	Insertions	Validation Loss	Mer	Substitutions	Wer	Wil	Wip
1.7038	5.0	14885	0.1731	6463	103703	7394	1.5150	0.3334	38015	0.3501	0.5133	0.4867
1.6051	6.0	17862	0.1512	6309	107805	6257	1.4257	0.3020	34067	0.3147	0.4705	0.5295
1.5396	7.0	20839	0.1368	5262	110900	6029	1.3715	0.2809	32019	0.2923	0.4428	0.5572
1.4436	8.0	23816	0.1299	5464	112866	5686	1.3285	0.2665	29851	0.2767	0.4207	0.5793
1.4287	9.0	26793	0.1241	4793	114249	5680	1.3090	0.2575	29139	0.2673	0.4091	0.5909
1.395	10.0	29770	0.1200	4698	115580	5505	1.2842	0.2479	27903	0.2572	0.3949	0.6051
1.3668	11.0	32747	0.1111	4400	117281	4991	1.2510	0.2343	26500	0.2422	0.3761	0.6239
1.3238	12.0	35724	0.1064	4560	117845	4500	1.2363	0.2282	25776	0.2351	0.3673	0.6327
1.3133	13.0	38701	0.1028	4173	118971	4613	1.2215	0.2214	25037	0.2283	0.3573	0.6427
1.2968	14.0	41678	0.0995	3937	119798	4466	1.2026	0.2152	24446	0.2217	0.3487	0.6513
1.2783	15.0	44655	0.0974	4071	120231	4295	1.1939	0.2115	23879	0.2176	0.3427	0.6573
1.2359	16.0	47632	0.0961	3946	120640	4313	1.1884	0.2089	23595	0.2150	0.3388	0.6612
1.2543	17.0	50609	0.0939	3757	121623	4476	1.1743	0.2033	22801	0.2094	0.3296	0.6704
1.2245	18.0	53586	0.0919	3981	121522	3944	1.1690	0.2012	22678	0.2065	0.3273	0.6727
1.2	19.0	56563	0.0903	3819	122029	3995	1.1626	0.1981	22333	0.2034	0.3226	0.6774
1.1964	20.0	59540	0.0916	3822	122170	4154	1.1598	0.1980	22189	0.2036	0.3218	0.6782
1.1822	21.0	62517	0.0871	3630	122825	3981	1.1471	0.1928	21726	0.1980	0.3146	0.6854
1.1758	22.0	65494	0.0862	3556	123114	3918	1.1413	0.1906	21511	0.1956	0.3114	0.6886
1.1735	23.0	68471	0.0847	3431	123623	4013	1.1381	0.1877	21127	0.1928	0.3067	0.6933
1.1556	24.0	71448	0.0839	3668	123854	3698	1.1282	0.1845	20659	0.1891	0.3015	0.6985
1.1538	25.0	74425	0.0819	3475	124201	3716	1.1240	0.1823	20505	0.1869	0.2986	0.7014
1.1078	26.0	77402	0.0819	3410	124426	3751	1.1259	0.1810	20345	0.1856	0.2965	0.7035
1.1539	27.0	80379	0.0805	3333	124879	3716	1.1152	0.1779	19969	0.1823	0.2916	0.7084
1.1432	54.0	80406	1.1113	0.0787	0.1790	0.1747	0.2868	0.7132	125277	19604	3300	3619
1.1171	55.0	81895	1.0912	0.0744	0.1713	0.1676	0.2763	0.7237	126048	18870	3263	3245
1.1027	56.0	83384	1.0874	0.0740	0.1696	0.1659	0.2736	0.7264	126362	18663	3156	3309
1.0827	57.0	84873	1.0865	0.0725	0.1690	0.1654	0.2728	0.7272	126356	18599	3226	3214
1.0794	58.0	86362	1.0837	0.0717	0.1665	0.1629	0.2691	0.7309	126790	18361	3030	3287
1.0585	59.0	87851	1.0816	0.0710	0.1664	0.1629	0.2688	0.7312	126738	18285	3158	3218
1.0549	60.0	89340	1.0785	0.0707	0.1651	0.1616	0.2671	0.7329	126913	18198	3070	3195
1.0708	61.0	90829	1.0795	0.0704	0.1649	0.1614	0.2667	0.7333	126928	18157	3096	3178
1.0674	62.0	92318	1.0767	0.0699	0.1638	0.1605	0.2650	0.7350	126981	17994	3206	3071
1.0709	63.0	93807	1.0738	0.0699	0.1638	0.1605	0.2652	0.7348	126999	18030	3152	3096
1.0672	64.0	95296	1.0734	0.0687	0.1622	0.1588	0.2630	0.7370	127257	17925	2999	3105
1.0716	65.0	96785	1.0712	0.0685	0.1610	0.1577	0.2613	0.7387	127412	17804	2965	3082
1.0664	66.0	98274	1.0723	0.0686	0.1613	0.1581	0.2618	0.7382	127312	17817	3052	3039
1.0452	67.0	99763	1.0703	0.0681	0.1605	0.1572	0.2605	0.7395	127444	17742	2995	3041
1.0318	68.0	101252	1.0695	0.0679	0.1603	0.1571	0.2601	0.7399	127479	17693	3009	3049
1.0341	69.0	102741	1.0686	0.0677	0.1597	0.1565	0.2590	0.7410	127600	17599	2982	3088
1.0338	70.0	104230	1.0688	0.0677	0.1598	0.1565	0.2593	0.7407	127573	17637	2971	3069

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.2.0+rocm5.6
• Datasets 2.18.0
• Tokenizers 0.15.2

Wandb run

https://wandb.ai/butspeechfit/decred_commonvoice_en/runs/ED_small_cv_v2_continue3