w2vbert-ctc-salt

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3020
• Wer: 0.3905
• Cer: 0.0840

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.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: 0.1
• training_steps: 15000

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
5.9489	0.2076	1500	1.0	3.0230	1.0
1.5319	0.4152	3000	0.5960	0.5589	0.1293
1.1602	0.6228	4500	0.4309	0.4809	0.1054
1.0148	0.8304	6000	0.3715	0.4499	0.0974
0.9507	1.0381	7500	0.3443	0.4274	0.0927
0.9469	1.2457	9000	0.3220	0.4031	0.0876
0.8564	1.4533	10500	0.3134	0.3995	0.0864
0.8318	1.6609	12000	0.3061	0.3951	0.0848
0.8707	1.8685	13500	0.3033	0.3904	0.0841
0.9274	2.0761	15000	0.3020	0.3905	0.0840

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu130
• Datasets 4.6.0
• Tokenizers 0.22.2