outputs

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: 0.3841
• Wer: 0.3473

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
8.3165	0.4235	500	8.7247	1.0
5.9446	0.8471	1000	5.9905	1.0
3.6231	1.2702	1500	3.6425	0.9965
3.2223	1.6938	2000	3.1964	1.0000
2.5537	2.1169	2500	2.5246	0.9821
1.7362	2.5404	3000	1.7241	0.7495
1.4488	2.9640	3500	1.4085	0.6282
1.3483	3.3871	4000	1.2128	0.5711
1.1602	3.8107	4500	1.0322	0.5253
0.9772	4.2338	5000	0.9011	0.4823
0.9321	4.6573	5500	0.7955	0.4555
0.7417	5.0805	6000	0.7076	0.4261
0.6725	5.5040	6500	0.6437	0.4170
0.5879	5.9276	7000	0.5842	0.4020
1.1923	6.3507	7500	0.5387	0.3902
0.6869	6.7742	8000	0.4979	0.3812
0.5258	7.1974	8500	0.4676	0.3731
0.5222	7.6209	9000	0.4455	0.3718
0.4546	8.0440	9500	0.4276	0.3647
0.5738	8.4676	10000	0.4065	0.3588
0.4969	8.8911	10500	0.3952	0.3534
0.4212	9.3143	11000	0.3898	0.3497
0.4783	9.7378	11500	0.3841	0.3473

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.22.1