linl03/whisper-small-vi

Whisper small vi - Ox

This model is a fine-tuned version of openai/whisper-small on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.0138
• Wer: 31.2667

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• 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
0.2276	0.08	1000	0.7506	29.8509
0.1768	0.16	2000	0.8114	31.2189
0.1828	0.24	3000	0.8569	31.2985
0.1632	0.32	4000	0.8523	31.9268
0.1566	0.4	5000	0.9062	31.9149
0.1532	0.48	6000	0.8914	31.4496
0.1593	0.56	7000	0.9342	31.9825
0.1411	0.64	8000	0.9412	32.0302
0.1531	0.72	9000	0.9456	31.6206
0.1246	0.8	10000	0.9452	31.7240
0.1336	0.88	11000	0.9622	31.1195
0.1392	0.96	12000	0.9638	31.3939
0.0725	1.04	13000	1.0032	31.5649
0.0838	1.12	14000	1.0346	31.7916
0.0766	1.2	15000	1.0138	31.2667

Framework versions

• Transformers 4.39.3
• Pytorch 2.4.1
• Datasets 3.0.1
• Tokenizers 0.15.2