wav2vec2-ft-datastf5-5lre5-adm-ga2b16-st30k-pat3

This model is a fine-tuned version of jonatasgrosman/wav2vec2-large-english on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.4189
• Wer: 0.6196
• Cer: 0.3498

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: 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_steps: 1500
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.2717	0.6200	500	3.1800	0.9996	0.9865
3.1348	1.2393	1000	3.1368	0.9996	0.9865
3.1281	1.8593	1500	3.1155	0.9996	0.9865
3.0486	2.4786	2000	3.0751	0.9996	0.9865
3.0129	3.0980	2500	3.0743	0.9996	0.9865
2.2751	3.7179	3000	1.9527	0.7681	0.4724
1.9867	4.3373	3500	1.8321	0.7165	0.4264
1.9045	4.9572	4000	1.7040	0.7002	0.4117
1.8447	5.5766	4500	1.6669	0.6877	0.4069
1.7401	6.1959	5000	1.6259	0.6771	0.3974
1.7741	6.8159	5500	1.6168	0.6671	0.3885
1.7329	7.4352	6000	1.5602	0.6625	0.3882
1.6152	8.0546	6500	1.5646	0.6543	0.3839
1.6342	8.6745	7000	1.5217	0.6537	0.3752
1.5794	9.2939	7500	1.4907	0.6491	0.3834
1.6278	9.9138	8000	1.4962	0.6497	0.3755
1.5393	10.5332	8500	1.4716	0.6493	0.3854
1.3962	11.1525	9000	1.4771	0.6328	0.3691
1.4951	11.7725	9500	1.4085	0.6310	0.3670
1.461	12.3918	10000	1.4111	0.6249	0.3619
1.4018	13.0112	10500	1.3925	0.6203	0.3574
1.4434	13.6311	11000	1.4463	0.6256	0.3569
1.2815	14.2505	11500	1.4169	0.6223	0.3593
1.3749	14.8704	12000	1.3655	0.6102	0.3478
1.3046	15.4898	12500	1.3809	0.6122	0.3513
1.252	16.1091	13000	1.3971	0.6100	0.3500
1.2503	16.7291	13500	1.4189	0.6196	0.3498

Framework versions

• Transformers 4.52.3
• Pytorch 2.7.0+cu118
• Datasets 3.6.0
• Tokenizers 0.21.1