outs

This model is a fine-tuned version of biodatlab/whisper-th-medium-combined on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0878
• Cer: 5.4469

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_steps: 5000
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.0275	0.9926	67	0.0804	4.9648
0.0359	2.0	135	0.0807	6.0688
0.0317	2.9926	202	0.0813	5.8626
0.0196	4.0	270	0.0819	4.9681
0.021	4.9926	337	0.0823	4.8683
0.0209	6.0	405	0.0831	4.8783
0.0223	6.9926	472	0.0866	5.0080
0.0173	8.0	540	0.0864	5.2541
0.0231	8.9926	607	0.0839	4.6422
0.0131	10.0	675	0.0860	5.3405
0.0134	10.9926	742	0.0899	5.2674
0.0183	12.0	810	0.0873	5.8094
0.0172	12.9926	877	0.0893	4.9215
0.0178	14.0	945	0.0855	5.7096
0.0127	14.8889	1005	0.0878	5.4469

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1