wav2vec2-large-xls-r-300m-gn

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1074
• Wer: 0.8952

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: 0.0003
• train_batch_size: 12
• eval_batch_size: 12
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 96
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4445	0.45	400	0.6193	0.6606
0.4459	0.91	800	0.3260	0.3907
0.3307	1.36	1200	0.2739	0.3315
0.3049	1.81	1600	0.2565	0.3027
0.2688	2.27	2000	0.2526	0.2863
0.2589	2.72	2400	0.2426	0.2821
0.2608	3.17	2800	0.2513	0.2965
0.2384	3.62	3200	0.2555	0.3052
0.2504	4.08	3600	0.2462	0.2855
0.2193	4.53	4000	0.2367	0.2691
0.2177	4.99	4400	0.2313	0.2637
0.2029	5.44	4800	0.2344	0.2633
0.2032	5.89	5200	0.2248	0.2553
0.1921	6.35	5600	0.2286	0.2668
0.188	6.8	6000	0.2239	0.2550
0.1808	7.25	6400	0.2323	0.2546
0.1791	7.71	6800	0.2285	0.2500
0.1796	8.16	7200	0.2467	0.2653
0.3112	8.61	7600	0.3921	0.3988
0.3545	9.07	8000	0.3703	0.3951
0.2528	9.52	8400	0.2441	0.2764
0.1932	9.97	8800	0.2385	0.2659
0.1688	10.42	9200	0.2245	0.2413
0.1645	10.88	9600	0.2220	0.2396
0.1736	11.34	10000	0.2456	0.2421
0.3031	11.79	10400	0.4347	0.2643
0.5795	12.24	10800	0.6777	0.3291
0.7227	12.7	11200	0.6537	0.3372
0.8282	13.15	11600	1.0544	0.8223
0.9917	13.6	12000	1.0290	0.7793
0.9407	14.06	12400	1.0095	0.7090
1.1154	14.51	12800	1.1014	0.8454
1.1354	14.97	13200	1.1074	0.8952

Framework versions

• Transformers 4.29.2
• Pytorch 2.0.1+cu117
• Datasets 2.12.0
• Tokenizers 0.13.3