nhi_heldout-speaker-exp_MJM502_mms-1b-nhi-adapterft

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: 0.7134
• Wer: 0.4967
• Cer: 0.1520

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: 16
• 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
1.0124	1.6393	200	0.9195	0.7361	0.2359
0.8011	3.2787	400	0.8468	0.7438	0.2192
0.7141	4.9180	600	0.7913	0.6398	0.2030
0.6767	6.5574	800	0.7461	0.6248	0.1987
0.629	8.1967	1000	0.7092	0.6241	0.1901
0.5964	9.8361	1200	0.7196	0.5959	0.1848
0.5531	11.4754	1400	0.6797	0.6012	0.1894
0.5541	13.1148	1600	0.6817	0.5659	0.1765
0.5166	14.7541	1800	0.6663	0.5588	0.1686
0.4807	16.3934	2000	0.6741	0.5567	0.1729
0.5044	18.0328	2200	0.6592	0.5455	0.1685
0.48	19.6721	2400	0.6720	0.5484	0.1698
0.467	21.3115	2600	0.6492	0.5479	0.1661
0.4634	22.9508	2800	0.6508	0.5430	0.1659
0.4503	24.5902	3000	0.6551	0.5521	0.1705
0.4373	26.2295	3200	0.6560	0.5328	0.1640
0.4139	27.8689	3400	0.6635	0.5354	0.1642
0.4093	29.5082	3600	0.6482	0.5310	0.1623
0.3878	31.1475	3800	0.6557	0.5158	0.1598
0.3967	32.7869	4000	0.6564	0.5253	0.1615
0.3724	34.4262	4200	0.6564	0.5121	0.1587
0.3791	36.0656	4400	0.6628	0.5172	0.1589
0.3608	37.7049	4600	0.6710	0.5231	0.1626
0.364	39.3443	4800	0.6523	0.5088	0.1575
0.3472	40.9836	5000	0.6687	0.5251	0.1623
0.3432	42.6230	5200	0.6622	0.5200	0.1598
0.3459	44.2623	5400	0.6517	0.5128	0.1569
0.3135	45.9016	5600	0.6543	0.5184	0.1606
0.3158	47.5410	5800	0.6571	0.5176	0.1590
0.3266	49.1803	6000	0.6657	0.5146	0.1567
0.2958	50.8197	6200	0.6676	0.5099	0.1563
0.2872	52.4590	6400	0.6734	0.5119	0.1572
0.2789	54.0984	6600	0.6743	0.5166	0.1575
0.287	55.7377	6800	0.6944	0.5111	0.1557
0.2911	57.3770	7000	0.6754	0.5052	0.1548
0.2782	59.0164	7200	0.6671	0.5095	0.1551
0.2694	60.6557	7400	0.6752	0.5040	0.1532
0.2598	62.2951	7600	0.6878	0.5113	0.1562
0.2688	63.9344	7800	0.6622	0.5064	0.1548
0.2633	65.5738	8000	0.6940	0.5075	0.1557
0.2454	67.2131	8200	0.6961	0.5025	0.1522
0.245	68.8525	8400	0.7007	0.5048	0.1540
0.24	70.4918	8600	0.6965	0.5088	0.1552
0.2456	72.1311	8800	0.6918	0.5092	0.1553
0.2429	73.7705	9000	0.7023	0.5075	0.1550
0.2283	75.4098	9200	0.7149	0.5084	0.1555
0.2236	77.0492	9400	0.7001	0.5083	0.1547
0.2299	78.6885	9600	0.6943	0.5035	0.1539
0.2184	80.3279	9800	0.7046	0.5068	0.1535
0.2142	81.9672	10000	0.6988	0.4985	0.1524
0.2123	83.6066	10200	0.7084	0.4987	0.1520
0.2197	85.2459	10400	0.7023	0.5002	0.1514
0.2012	86.8852	10600	0.7108	0.5024	0.1526
0.2079	88.5246	10800	0.7081	0.4986	0.1525
0.209	90.1639	11000	0.7056	0.5047	0.1533
0.1943	91.8033	11200	0.7143	0.5005	0.1518
0.1965	93.4426	11400	0.7077	0.4988	0.1522
0.1886	95.0820	11600	0.7179	0.4960	0.1516
0.2085	96.7213	11800	0.7170	0.4979	0.1520
0.1812	98.3607	12000	0.7129	0.4978	0.1523
0.1904	100.0	12200	0.7134	0.4967	0.1520

Framework versions

• Transformers 4.41.2
• Pytorch 2.4.0
• Datasets 3.2.0
• Tokenizers 0.19.1