dataprizma/whisper-medium-uz

Whisper Medium UZ - Bahriddin Mo'minov

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 16.1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2593
• Wer: 17.2801

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4581	0.2641	1000	0.3953	30.3663
0.3859	0.5282	2000	0.3257	26.6366
0.2601	0.7923	3000	0.2745	22.5373
0.1724	1.0564	4000	0.2611	21.4740
0.1338	1.3205	5000	0.2526	20.7058
0.1434	1.5846	6000	0.2428	19.7434
0.1136	1.8487	7000	0.2362	19.1380
0.0783	2.1128	8000	0.2387	18.7193
0.0692	2.3769	9000	0.2349	18.4846
0.0722	2.6410	10000	0.2343	18.8605
0.0683	2.9051	11000	0.2297	18.0129
0.0482	3.1692	12000	0.2443	18.1920
0.0231	3.4332	13000	0.2442	17.7089
0.0255	3.6973	14000	0.2468	17.7821
0.022	3.9614	15000	0.2455	17.5538
0.0092	4.2255	16000	0.2553	17.5424
0.0058	4.4896	17000	0.2614	17.5828
0.0048	4.7537	18000	0.2593	17.2801

Framework versions

• Transformers 4.41.1
• Pytorch 2.2.2+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1