w2v-V3

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1847
• Wer: 0.1613

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.566	0.0428	300	0.6246	0.5686
0.462	0.0856	600	0.5791	0.3623
0.4407	0.1284	900	0.4428	0.3232
0.4036	0.1712	1200	0.4119	0.3066
0.328	0.2139	1500	0.3693	0.2684
0.3151	0.2567	1800	0.3102	0.2462
0.2907	0.2995	2100	0.3221	0.2411
0.2553	0.3423	2400	0.3061	0.2430
0.2156	0.3851	2700	0.2857	0.2104
0.2034	0.4279	3000	0.2516	0.2025
0.2038	0.4707	3300	0.2395	0.1995
0.1751	0.5135	3600	0.2372	0.1875
0.1697	0.5563	3900	0.2063	0.1809
0.1501	0.5991	4200	0.2005	0.1775
0.1428	0.6418	4500	0.2024	0.1701
0.1211	0.6846	4800	0.1883	0.1642

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0