wav2vec2-large-mms-1b-nhi-adapterft-orig-ortho_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: 0.6127
• Wer: 0.4134
• Cer: 0.1235

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
1.0469	1.6529	200	0.7880	0.6456	0.2068
0.8396	3.3058	400	0.6866	0.6118	0.1870
0.7648	4.9587	600	0.6500	0.5656	0.1752
0.71	6.6116	800	0.6277	0.5477	0.1686
0.6709	8.2645	1000	0.6158	0.5349	0.1585
0.6492	9.9174	1200	0.5821	0.5033	0.1501
0.6284	11.5702	1400	0.5741	0.5145	0.1573
0.6054	13.2231	1600	0.5814	0.4912	0.1494
0.5808	14.8760	1800	0.5456	0.4896	0.1444
0.5585	16.5289	2000	0.5647	0.4884	0.1465
0.5546	18.1818	2200	0.5639	0.4890	0.1475
0.5288	19.8347	2400	0.5573	0.4903	0.1475
0.5307	21.4876	2600	0.5480	0.4657	0.1397
0.5172	23.1405	2800	0.5427	0.4622	0.1386
0.5041	24.7934	3000	0.5377	0.4549	0.1344
0.4748	26.4463	3200	0.5483	0.4635	0.1370
0.4754	28.0992	3400	0.5447	0.4699	0.1429
0.4602	29.7521	3600	0.5495	0.4523	0.1353
0.4502	31.4050	3800	0.5457	0.4329	0.1286
0.4413	33.0579	4000	0.5515	0.4501	0.1325
0.4391	34.7107	4200	0.5263	0.4545	0.1320
0.4097	36.3636	4400	0.5485	0.4574	0.1365
0.4208	38.0165	4600	0.5394	0.4542	0.1336
0.4086	39.6694	4800	0.5392	0.4357	0.1294
0.3956	41.3223	5000	0.5579	0.4332	0.1304
0.4036	42.9752	5200	0.5475	0.4376	0.1307
0.3984	44.6281	5400	0.5492	0.4297	0.1295
0.3769	46.2810	5600	0.5503	0.4348	0.1289
0.3699	47.9339	5800	0.5330	0.4357	0.1284
0.3611	49.5868	6000	0.5682	0.4380	0.1308
0.3619	51.2397	6200	0.5661	0.4316	0.1276
0.3387	52.8926	6400	0.5512	0.4287	0.1282
0.3392	54.5455	6600	0.5834	0.4351	0.1291
0.3365	56.1983	6800	0.5710	0.4335	0.1276
0.3288	57.8512	7000	0.5631	0.4262	0.1287
0.3244	59.5041	7200	0.5605	0.4281	0.1272
0.3187	61.1570	7400	0.5695	0.4332	0.1275
0.3258	62.8099	7600	0.5684	0.4265	0.1268
0.3035	64.4628	7800	0.5924	0.4185	0.1254
0.3051	66.1157	8000	0.5732	0.4319	0.1279
0.2968	67.7686	8200	0.5773	0.4204	0.1249
0.2982	69.4215	8400	0.5819	0.4140	0.1243
0.297	71.0744	8600	0.5941	0.4159	0.1240
0.2922	72.7273	8800	0.5836	0.4201	0.1229
0.2798	74.3802	9000	0.5951	0.4201	0.1243
0.2692	76.0331	9200	0.5820	0.4220	0.1255
0.2704	77.6860	9400	0.5954	0.4230	0.1251
0.271	79.3388	9600	0.6022	0.4172	0.1254
0.2633	80.9917	9800	0.5975	0.4182	0.1248
0.2554	82.6446	10000	0.6114	0.4124	0.1242
0.2575	84.2975	10200	0.6084	0.4153	0.1235
0.2554	85.9504	10400	0.6007	0.4156	0.1243
0.2595	87.6033	10600	0.6010	0.4166	0.1240
0.2544	89.2562	10800	0.6080	0.4217	0.1251
0.2555	90.9091	11000	0.6076	0.4156	0.1246
0.247	92.5620	11200	0.6151	0.4150	0.1239
0.2465	94.2149	11400	0.6113	0.4121	0.1240
0.2376	95.8678	11600	0.6136	0.4153	0.1237
0.2464	97.5207	11800	0.6121	0.4137	0.1235
0.2433	99.1736	12000	0.6127	0.4134	0.1235

Framework versions

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