DewiBrynJones/wav2vec2-xlsr-53-ft-cy-2601

wav2vec2-xlsr-53-ft-btb-cv-cvad-cven-wlga-ca

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2807
• Wer: 0.2338

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• training_steps: 18000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9448	0.1188	1000	0.8082	0.5919
0.7677	0.2375	2000	0.5489	0.4368
0.683	0.3563	3000	0.4794	0.3883
0.6549	0.4751	4000	0.4609	0.3699
0.6388	0.5939	5000	0.4409	0.3651
0.6161	0.7126	6000	0.4129	0.3377
0.5878	0.8314	7000	0.3809	0.3200
0.5527	0.9502	8000	0.3641	0.3010
0.5052	1.0689	9000	0.3495	0.2888
0.4827	1.1877	10000	0.3397	0.2788
0.4698	1.3064	11000	0.3247	0.2685
0.4603	1.4252	12000	0.3081	0.2585
0.4513	1.5440	13000	0.3030	0.2511
0.4335	1.6627	14000	0.2905	0.2426
0.4196	1.7815	15000	0.2873	0.2392
0.4213	1.9003	16000	0.2800	0.2355
0.3709	2.0190	17000	0.2804	0.2339
0.4117	2.1378	18000	0.2807	0.2338

Framework versions

• Transformers 4.57.6
• Pytorch 2.9.1+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2