wav2vec2-large-xls-r-300m-af-1hr-v2

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: 1.4465
• Wer: 0.6609
• Cer: 0.2264

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
13.6642	2.7778	100	6.3667	1.0	1.0
4.9297	5.5556	200	3.8947	1.0	1.0
3.3634	8.3333	300	3.0087	1.0	1.0
2.9415	11.1111	400	2.9310	1.0	1.0
2.8979	13.8889	500	2.8946	1.0	1.0
2.8804	16.6667	600	2.8853	1.0	1.0
2.8089	19.4444	700	2.5587	1.0	0.9801
1.5299	22.2222	800	1.3499	0.8436	0.2927
0.6494	25.0	900	1.2093	0.6976	0.2505
0.3557	27.7778	1000	1.2005	0.6841	0.2389
0.249	30.5556	1100	1.2462	0.6865	0.2371
0.1929	33.3333	1200	1.3202	0.6744	0.2335
0.1545	36.1111	1300	1.3346	0.6676	0.2282
0.1352	38.8889	1400	1.4056	0.6768	0.2361
0.1182	41.6667	1500	1.3930	0.6635	0.2326
0.1063	44.4444	1600	1.4446	0.6573	0.2293
0.0972	47.2222	1700	1.3908	0.6531	0.2273
0.0804	50.0	1800	1.4328	0.6541	0.2262
0.0816	52.7778	1900	1.4489	0.6586	0.2263
0.078	55.5556	2000	1.4210	0.6581	0.2268
0.0714	58.3333	2100	1.4393	0.6581	0.2260

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1