pother/whisper-th-small-bio-finetuned

Whisper Small Th Combined Finetuned

This model is a fine-tuned version of biodatlab/whisper-th-small-combined on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0702
• Wer: 0.4132

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3362	0.2175	1000	0.1439	0.6061
0.2993	0.4349	2000	0.1230	0.5645
0.2523	0.6524	3000	0.1080	0.5299
0.2823	0.8698	4000	0.0939	0.4914
0.2459	1.0873	5000	0.0840	0.4570
0.2005	1.3047	6000	0.0776	0.4364
0.2081	1.5222	7000	0.0724	0.4157
0.1918	1.7396	8000	0.0702	0.4132

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0