anuragshas/wav2vec2-xls-r-300m-mr-cv9-with-lm

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_9_0 - MR dataset. It achieves the following results on the evaluation set:

• Loss: 0.3642
• Wer: 0.4190
• Cer: 0.0946

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: 7.5e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 6124
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.5184	12.9	400	3.4210	1.0	1.0
2.3797	25.81	800	1.1068	0.8389	0.2584
1.5022	38.71	1200	0.5278	0.6280	0.1517
1.3181	51.61	1600	0.4254	0.5587	0.1297
1.2037	64.52	2000	0.3836	0.5143	0.1176
1.1245	77.42	2400	0.3643	0.4871	0.1111
1.0582	90.32	2800	0.3562	0.4676	0.1062
1.0027	103.23	3200	0.3530	0.4625	0.1058
0.9382	116.13	3600	0.3388	0.4442	0.1002
0.8915	129.03	4000	0.3430	0.4427	0.1000
0.853	141.94	4400	0.3536	0.4375	0.1000
0.8127	154.84	4800	0.3511	0.4344	0.0986
0.7861	167.74	5200	0.3595	0.4372	0.0993
0.7619	180.65	5600	0.3628	0.4316	0.0985
0.7537	193.55	6000	0.3633	0.4174	0.0943

Framework versions

• Transformers 4.19.0.dev0
• Pytorch 1.11.0+cu102
• Datasets 2.1.1.dev0
• Tokenizers 0.12.1