rngzhi/cs3264-project-v2

Whisper Small - Singlish v2

This model is a fine-tuned version of openai/whisper-small on the rngzhi/cs3264-project dataset. It achieves the following results on the evaluation set:

• Loss: 0.1850
• Wer: 4.9236

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: 16
• eval_batch_size: 8
• 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: linear
• lr_scheduler_warmup_steps: 25
• training_steps: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5404	0.0625	50	0.1970	5.6075
0.075	1.0144	100	0.1557	4.8780
0.042	1.0769	150	0.1610	4.9692
0.0185	2.0288	200	0.1628	4.9122
0.0117	2.0913	250	0.1651	5.0262
0.0096	3.0431	300	0.1716	5.0490
0.007	3.1056	350	0.1747	5.0034
0.0045	4.0575	400	0.1783	5.1402
0.0046	5.0094	450	0.1749	5.1288
0.004	5.0719	500	0.1782	5.0148
0.0021	6.0237	550	0.1814	5.0034
0.004	6.0862	600	0.1813	4.9920
0.0024	7.0381	650	0.1844	4.9350
0.0022	7.1006	700	0.1834	4.9008
0.0032	8.0525	750	0.1850	4.9236
0.0016	9.0044	800	0.1850	4.9236

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.2.1+cu121
• Datasets 2.18.1.dev0
• Tokenizers 0.19.1