w2v-bert-bem-natbed-combined-model

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

• Loss: 0.6289
• Wer: 0.6078

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: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0078	0.5006	200	0.9815	0.8201
0.8769	1.0013	400	0.9823	1.0433
0.805	1.5019	600	0.8306	0.8606
0.8141	2.0025	800	0.7548	0.7196
0.7132	2.5031	1000	0.7485	0.6932
0.7058	3.0038	1200	0.7280	0.6917
0.6563	3.5044	1400	0.7046	0.7045
0.6232	4.0050	1600	0.7186	0.7409
0.6093	4.5056	1800	0.7048	0.6434
0.5767	5.0063	2000	0.6521	0.6474
0.5628	5.5069	2200	0.6322	0.6018
0.5569	6.0075	2400	0.6289	0.6078
0.5156	6.5081	2600	0.6504	0.6374
0.5074	7.0088	2800	0.6638	0.6222
0.4906	7.5094	3000	0.6744	0.5884

Framework versions

• Transformers 4.46.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0