wav2vec2-large-mms-1b-nhi-adapterft-ilv_fold1

This model is a fine-tuned version of facebook/mms-1b-all on the audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.6054
• Wer: 0.3871
• Cer: 0.1165

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: 20
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.312	1.6529	200	0.8322	0.7050	0.2128
0.8371	3.3058	400	0.6980	0.5965	0.1790
0.7485	4.9587	600	0.6473	0.5388	0.1606
0.6799	6.6116	800	0.6305	0.5407	0.1561
0.6308	8.2645	1000	0.6215	0.4876	0.1467
0.6359	9.9174	1200	0.6104	0.4979	0.1457
0.5936	11.5702	1400	0.5851	0.4799	0.1431
0.5757	13.2231	1600	0.5906	0.4769	0.1389
0.5479	14.8760	1800	0.5851	0.4826	0.1377
0.5516	16.5289	2000	0.5596	0.4685	0.1373
0.5269	18.1818	2200	0.5788	0.4624	0.1334
0.5131	19.8347	2400	0.5708	0.4952	0.1411
0.495	21.4876	2600	0.5743	0.4532	0.1303
0.4898	23.1405	2800	0.5731	0.4421	0.1290
0.4718	24.7934	3000	0.5610	0.4471	0.1307
0.4717	26.4463	3200	0.5691	0.4498	0.1307
0.4573	28.0992	3400	0.5787	0.4348	0.1277
0.4587	29.7521	3600	0.5597	0.4345	0.1253
0.4265	31.4050	3800	0.5582	0.4219	0.1246
0.4383	33.0579	4000	0.5665	0.4127	0.1229
0.416	34.7107	4200	0.5615	0.4241	0.1256
0.408	36.3636	4400	0.5561	0.4264	0.1257
0.4014	38.0165	4600	0.5676	0.4272	0.1255
0.3929	39.6694	4800	0.5799	0.4199	0.1224
0.3764	41.3223	5000	0.5859	0.4119	0.1197
0.3765	42.9752	5200	0.5763	0.4092	0.1207
0.3783	44.6281	5400	0.5926	0.4264	0.1241
0.36	46.2810	5600	0.5598	0.4012	0.1182
0.3625	47.9339	5800	0.5811	0.4257	0.1228
0.3615	49.5868	6000	0.5826	0.4096	0.1215
0.3331	51.2397	6200	0.5792	0.4119	0.1211
0.3378	52.8926	6400	0.5749	0.3993	0.1192
0.3322	54.5455	6600	0.5691	0.4100	0.1224
0.3298	56.1983	6800	0.5781	0.4020	0.1188
0.3216	57.8512	7000	0.5834	0.4131	0.1207
0.3141	59.5041	7200	0.5874	0.3963	0.1184
0.303	61.1570	7400	0.5974	0.3966	0.1161
0.3128	62.8099	7600	0.5845	0.3982	0.1169
0.2936	64.4628	7800	0.5694	0.4043	0.1172
0.2951	66.1157	8000	0.5751	0.4054	0.1195
0.2889	67.7686	8200	0.6028	0.4035	0.1189
0.2735	69.4215	8400	0.5818	0.3951	0.1169
0.2858	71.0744	8600	0.6124	0.3924	0.1184
0.2707	72.7273	8800	0.5837	0.3863	0.1147
0.2788	74.3802	9000	0.5813	0.3943	0.1168
0.2649	76.0331	9200	0.5954	0.3921	0.1171
0.2606	77.6860	9400	0.5994	0.3898	0.1161
0.2698	79.3388	9600	0.5999	0.3921	0.1164
0.2529	80.9917	9800	0.5921	0.3917	0.1171
0.2504	82.6446	10000	0.5949	0.3901	0.1168
0.2618	84.2975	10200	0.5993	0.3970	0.1181
0.2549	85.9504	10400	0.6138	0.3936	0.1170
0.2529	87.6033	10600	0.6049	0.3890	0.1165
0.2448	89.2562	10800	0.6051	0.3882	0.1164
0.2461	90.9091	11000	0.6089	0.3913	0.1169
0.245	92.5620	11200	0.6056	0.3878	0.1172
0.2423	94.2149	11400	0.6094	0.3821	0.1151
0.2423	95.8678	11600	0.6002	0.3863	0.1151
0.2364	97.5207	11800	0.6006	0.3844	0.1155
0.2427	99.1736	12000	0.6054	0.3871	0.1165

Framework versions

• Transformers 4.41.2
• Pytorch 2.4.0
• Datasets 2.19.1
• Tokenizers 0.19.1