whisper-tiny-khmer-aug-v4

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

• Loss: 0.4707
• Wer: 87.9034

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: 0.0001
• 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: constant
• lr_scheduler_warmup_steps: 1000
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1549	1.0	793	1.0460	101.2810
0.8009	2.0	1586	0.7930	108.9509
0.5929	3.0	2379	0.5986	95.7516
0.4686	4.0	3172	0.5323	99.9351
0.411	5.0	3965	0.4952	88.9087
0.3694	6.0	4758	0.4675	101.5405
0.332	7.0	5551	0.4639	90.0600
0.3043	8.0	6344	0.4541	78.9363
0.2819	9.0	7137	0.4532	82.9901
0.2615	10.0	7930	0.4707	87.9034

Framework versions

• Transformers 4.44.0
• Pytorch 2.3.1
• Datasets 2.21.0
• Tokenizers 0.19.1