w2vbert-waxal

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: 0.8908
• Wer: 0.4034
• Cer: 0.2183
• Combined Err: 0.3108

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 300
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Combined Err
1.2784	0.0891	400	1.1531	0.5282	0.2566	0.3924
0.993	0.1782	800	1.0895	0.5016	0.2428	0.3722
1.0168	0.2674	1200	1.0470	0.4556	0.2395	0.3475
0.8664	0.3565	1600	1.0232	0.4507	0.2286	0.3397
0.8443	0.4456	2000	1.0147	0.4446	0.2309	0.3378
0.7291	0.5347	2400	0.9803	0.4406	0.2356	0.3381
0.7973	0.6239	2800	1.0150	0.4508	0.2221	0.3364
0.7143	0.7130	3200	1.0399	0.4021	0.2227	0.3124
0.8128	0.8021	3600	1.0372	0.4084	0.2242	0.3163
0.7897	0.8912	4000	0.9631	0.4180	0.2281	0.3230
0.6778	0.9803	4400	0.9587	0.4108	0.2217	0.3163
0.731	1.0693	4800	0.9741	0.4030	0.2230	0.3130
0.6586	1.1584	5200	0.9550	0.4450	0.2190	0.3320
0.7972	1.2475	5600	0.9066	0.4041	0.2207	0.3124
0.6907	1.3367	6000	0.8938	0.4598	0.2255	0.3426
0.7299	1.4258	6400	0.9437	0.4032	0.2211	0.3121
0.6452	1.5149	6800	0.9042	0.4144	0.2195	0.3169
0.7014	1.6040	7200	0.9247	0.3883	0.2143	0.3013
0.6198	1.6931	7600	0.9282	0.3988	0.2142	0.3065
0.8231	1.7823	8000	0.8880	0.4122	0.2178	0.3150
0.5864	1.8714	8400	0.8961	0.4032	0.2166	0.3099
0.744	1.9605	8800	0.8908	0.4034	0.2183	0.3108

Framework versions

• Transformers 4.53.2
• Pytorch 2.12.0+cu130
• Datasets 3.6.0
• Tokenizers 0.21.4