w2v-bert-2.0-sq

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4195
• Wer: 0.3544

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5
• num_epochs: 7
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.4885	0.6061	20	3.2689	1.0
3.0283	1.2121	40	3.0235	0.9949
1.7144	1.8182	60	1.3367	0.9483
0.6599	2.4242	80	0.7279	0.6317
0.6135	3.0303	100	0.6208	0.5615
0.4033	3.6364	120	0.5120	0.4730
0.2658	4.2424	140	0.4693	0.4270
0.3056	4.8485	160	0.4831	0.4327
0.2024	5.4545	180	0.4536	0.3991
0.1963	6.0606	200	0.4297	0.3747
0.1494	6.6667	220	0.4195	0.3544

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1