w2v-bert-2.0-luo_cv_fleurs_19h-v3

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the CLEAR-GLOBAL/LUO_19H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.2850
• Wer: 0.3100
• Cer: 0.0950

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: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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
• lr_scheduler_warmup_ratio: 0.05
• training_steps: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.9185	6.4935	1000	0.8056	0.6817	0.2146
0.3338	12.9870	2000	0.4123	0.4070	0.1262
0.1888	19.4805	3000	0.3017	0.3492	0.1042
0.1032	25.9740	4000	0.2851	0.3100	0.0946
0.0541	32.4675	5000	0.3172	0.3060	0.0946
0.0284	38.9610	6000	0.3164	0.2897	0.0912
0.0145	45.4545	7000	0.3478	0.2879	0.0904
0.0176	51.9481	8000	0.3971	0.3113	0.0933
0.0051	58.4416	9000	0.4149	0.3051	0.0905

Framework versions

• Transformers 4.48.1
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1