wav2vec2-large-mms-1b-tig-colab

This model is a fine-tuned version of facebook/mms-1b-all on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.9701
• Wer: 0.4444

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: 8
• 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
16.3491	3.4483	100	1.2238	0.8160
0.9459	6.8966	200	0.7144	0.5474
0.6542	10.3448	300	0.6971	0.5064
0.5291	13.7931	400	0.7007	0.5091
0.4158	17.2414	500	0.7214	0.4973
0.3521	20.6897	600	0.7371	0.4836
0.3123	24.1379	700	0.7818	0.4781
0.254	27.5862	800	0.7690	0.4718
0.2332	31.0345	900	0.8010	0.4663
0.2045	34.4828	1000	0.8207	0.4545
0.1903	37.9310	1100	0.8558	0.4636
0.1687	41.3793	1200	0.8713	0.4563
0.1604	44.8276	1300	0.8949	0.4791
0.1541	48.2759	1400	0.8712	0.4572
0.1291	51.7241	1500	0.8975	0.4417
0.1342	55.1724	1600	0.8952	0.4608
0.134	58.6207	1700	0.9179	0.4599
0.1245	62.0690	1800	0.9422	0.4599
0.1176	65.5172	1900	0.9493	0.4599
0.1144	68.9655	2000	0.9596	0.4454
0.1109	72.4138	2100	0.9491	0.4399
0.0995	75.8621	2200	0.9719	0.4517
0.091	79.3103	2300	0.9690	0.4617
0.0914	82.7586	2400	0.9767	0.4454
0.0964	86.2069	2500	0.9628	0.4499
0.09	89.6552	2600	0.9696	0.4490
0.0923	93.1034	2700	0.9653	0.4472
0.0921	96.5517	2800	0.9729	0.4399
0.0879	100.0	2900	0.9701	0.4444

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1