DeCRED_small_cv_v2_linear_mixing

This model is a fine-tuned version of Lakoc/DeCRED_small_cv_2 on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.9542
• Cer: 0.3765
• Wer: 0.6117
• Mer: 0.5575
• Wil: 0.7685
• Wip: 0.2315
• Hits: 22590
• Substitutions: 20263
• Deletions: 3668
• Insertions: 4527

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: 0.0005
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• 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
• num_epochs: 50.0

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer	Mer	Wil	Wip	Hits	Substitutions	Deletions	Insertions
6.9661	0.98	22	6.8477	60.1841	50.8357	0.9996	1.0000	0.0000	983	45533	5	2319391
6.5874	2.0	45	6.6100	59.9264	50.2003	0.9995	1.0000	0.0000	1108	45404	9	2289956
6.4843	2.98	67	6.3899	59.4350	49.5502	0.9995	1.0000	0.0000	1246	45261	14	2259851
6.1871	4.0	90	6.1667	58.7677	48.7290	0.9994	1.0000	0.0000	1390	45122	9	2221790
6.1088	4.98	112	5.9594	57.9917	47.9251	0.9993	1.0000	0.0000	1603	44900	18	2184606
5.8041	6.0	135	5.7487	56.6337	46.6790	0.9992	1.0000	0.0000	1816	44680	25	2126851
5.7494	6.98	157	5.5529	54.6535	44.8725	0.9991	1.0000	0.0000	1974	44513	34	2042966
5.4083	8.0	180	5.3546	52.4198	42.9372	0.9989	0.9999	0.0001	2173	44298	50	1953135
5.3779	8.98	202	5.1706	49.7925	40.7569	0.9988	0.9999	0.0001	2371	44091	59	1851904
5.02	10.0	225	4.9842	46.7020	38.1816	0.9985	0.9999	0.0001	2603	43831	87	1732326
4.9776	10.98	247	4.8119	43.6679	35.8707	0.9983	0.9999	0.0001	2852	43570	99	1625071
4.7425	12.0	270	4.6377	39.7527	32.6943	0.9980	0.9999	0.0001	3054	43352	115	1477503
4.608	12.98	292	4.4773	35.2066	29.0084	0.9976	0.9998	0.0002	3233	43132	156	1306210
4.4031	14.0	315	4.3150	31.5887	26.0092	0.9971	0.9998	0.0002	3487	42835	199	1166942
4.3239	14.98	337	4.1657	27.0209	22.5064	0.9965	0.9997	0.0003	3717	42481	323	1004215
4.1256	16.0	360	4.0154	22.4586	18.8399	0.9956	0.9996	0.0004	3907	42238	376	833835
4.0373	16.98	382	3.8773	18.2020	15.3318	0.9942	0.9995	0.0005	4128	41849	544	670857
3.8293	18.0	405	3.7389	14.5637	12.3297	0.9923	0.9993	0.0007	4442	41435	644	531510
3.7401	18.98	427	3.6120	11.4548	9.7572	0.9897	0.9990	0.0010	4708	41051	762	412101
3.5255	20.0	450	3.4851	8.4210	7.3279	0.9852	0.9984	0.0016	5122	40427	972	299500
3.5611	20.98	472	3.3694	5.8830	5.3130	0.9783	0.9974	0.0026	5473	39918	1130	206120
3.3464	22.0	495	3.2537	4.1319	3.8709	0.9682	0.9959	0.0041	5905	39233	1383	139463
3.3134	22.98	517	3.1489	3.1610	3.0400	0.9567	0.9940	0.0060	6408	38514	1599	101309
3.1154	24.0	540	3.0447	2.2506	2.2882	0.9392	0.9909	0.0091	6887	37758	1876	66816
3.0684	24.98	562	2.9503	1.5946	1.7552	0.9158	0.9861	0.0139	7503	36824	2194	42636
2.9926	26.0	585	2.8569	1.2290	1.4535	0.8931	0.9808	0.0192	8097	36034	2390	29195
2.9429	26.98	607	2.7728	1.0860	1.3147	0.8752	0.9757	0.0243	8722	35139	2660	23363
2.8033	28.0	630	2.6900	0.8996	1.1624	0.8519	0.9687	0.0313	9399	34374	2748	16952
2.7652	28.98	652	2.6158	0.8134	1.0854	0.8326	0.9615	0.0385	10155	33342	3024	14126
2.6598	30.0	675	2.5430	0.7254	1.0033	0.8098	0.9526	0.0474	10964	32379	3178	11116
2.6088	30.98	697	2.4781	0.6766	0.9584	0.7914	0.9439	0.0561	11755	31417	3349	9819
2.5442	32.0	720	2.4151	0.6563	0.9343	0.7759	0.9354	0.0646	12556	30412	3553	9501
2.5035	32.98	742	2.3592	0.6205	0.8964	0.7572	0.9252	0.0748	13370	29435	3716	8552
2.4259	34.0	765	2.3051	0.5803	0.8567	0.7354	0.9123	0.0877	14341	28415	3765	7674
2.3946	34.98	787	2.2576	0.5549	0.8295	0.7172	0.9004	0.0996	15216	27479	3826	7282
2.3014	36.0	810	2.2121	0.5257	0.8003	0.6971	0.8864	0.1136	16180	26476	3865	6888
2.2883	36.98	832	2.1725	0.5050	0.7753	0.6790	0.8733	0.1267	17049	25677	3795	6598
2.2694	38.0	855	2.1350	0.4803	0.7461	0.6596	0.8587	0.1413	17913	24794	3814	6102
2.2372	38.98	877	2.1028	0.4635	0.7254	0.6447	0.8465	0.1535	18597	24029	3895	5821
2.1639	40.0	900	2.0728	0.4458	0.7033	0.6289	0.8335	0.1665	19309	23310	3902	5508
2.1478	40.98	922	2.0475	0.4303	0.6843	0.6146	0.8211	0.1789	19960	22639	3922	5271
2.1546	42.0	945	2.0245	0.4172	0.6653	0.5999	0.8083	0.1917	20644	22064	3813	5075
2.1382	42.98	967	2.0056	0.4062	0.6510	0.5885	0.7979	0.2021	21179	21588	3754	4942
2.1007	44.0	990	1.9892	0.3961	0.6376	0.5780	0.7881	0.2119	21656	21111	3754	4798
2.09	44.98	1012	1.9766	0.3885	0.6282	0.5705	0.7810	0.2190	21996	20801	3724	4698
2.1065	46.0	1035	1.9664	0.3827	0.6207	0.5644	0.7752	0.2248	22286	20556	3679	4641
2.1115	46.98	1057	1.9596	0.3793	0.6157	0.5604	0.7713	0.2287	22466	20393	3662	4587
2.0602	48.0	1080	1.9554	0.3770	0.6125	0.5581	0.7691	0.2309	22564	20295	3662	4537
1.9657	48.89	1100	1.9542	0.3765	0.6117	0.5575	0.7685	0.2315	22590	20263	3668	4527

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/DeCRED_small_cv_v2_linear_mixing