MagiBoss/Whisper-Large-V3-Thai-Lora

Whisper Large V3 Thai Lora - Magi Boss

This model is a fine-tuned version of openai/whisper-large-v3 on the Common Voice 11.0 dataset (Training Set 20000 row, Validation Set 500 row). It achieves the following results on the evaluation set:

• Loss: 0.1894
• WER: 0.8101
• TER: 81.0089
• ChrF: 87.4811
• CER: 0.1041
• BLEU: 8.7391
• SubER: 0.8189

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 25
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Ter	Chrf	Cer	Bleu	SubER
0.3053	0.32	50	0.2052	0.8457	84.5697	86.6971	0.1115	7.8703	0.8640
0.3752	0.64	100	0.1937	0.8323	83.2344	86.9801	0.1087	8.1510	0.8469
0.2794	0.96	150	0.1894	0.8101	81.0089	87.4811	0.1041	8.7391	0.8189

Framework versions

• PEFT 0.12.1.dev0
• Transformers 4.45.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1