wav2vec2-large-mms-1b-vi-kag

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

• Loss: 0.2484
• Wer: 0.2168

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4347	0.2744	200	0.3215	0.2742
0.3992	0.5489	400	0.2968	0.2666
0.3398	0.8233	600	0.3041	0.2686
0.3506	1.0978	800	0.2813	0.2441
0.3331	1.3722	1000	0.2748	0.2394
0.3199	1.6467	1200	0.2738	0.2410
0.3153	1.9211	1400	0.2669	0.2344
0.3118	2.1955	1600	0.2612	0.2291
0.3059	2.4700	1800	0.2577	0.2313
0.2986	2.7444	2000	0.2570	0.2230
0.315	3.0189	2200	0.2543	0.2218
0.2949	3.2933	2400	0.2551	0.2239
0.2967	3.5678	2600	0.2497	0.2179
0.2956	3.8422	2800	0.2484	0.2168

Framework versions

• Transformers 4.41.2
• Pytorch 2.1.2+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1