manuth/large_v3_turbo

Whisper Based Finetuned Khmer

This model is a fine-tuned version of openai/whisper-base on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1346
• Wer: 0.5070

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 500
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5914	0.3170	200	0.5412	0.9067
0.3671	0.6339	400	0.3620	0.7633
0.2896	0.9509	600	0.3216	0.7284
0.2151	1.2678	800	0.2727	0.6579
0.1875	1.5848	1000	0.2191	0.6095
0.1708	1.9017	1200	0.1818	0.5762
0.1111	2.2187	1400	0.1728	0.5538
0.1017	2.5357	1600	0.1540	0.5296
0.0881	2.8526	1800	0.1346	0.5070

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0