whisper-base-khmer-aug

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

• Loss: 0.4161
• Wer: 73.7474

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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
1.0144	0.9989	669	0.6891	100.4865
0.5319	1.9993	1339	0.4952	89.9303
0.4108	2.9996	2009	0.4335	88.0979
0.3458	4.0	2679	0.4046	79.3903
0.3025	4.9989	3348	0.3869	76.9256
0.2688	5.9993	4018	0.3794	81.5145
0.2427	6.9996	4688	0.3875	77.5904
0.2187	8.0	5358	0.3883	74.6392
0.201	8.9989	6027	0.4046	74.5581
0.182	9.9888	6690	0.4161	73.7474

Framework versions

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