whisper-uzbek-multi-dataset

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2215
• Wer: 24.5188

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4219	0.1577	1000	0.4203	37.8390
0.3467	0.3154	2000	0.3568	35.5245
0.3152	0.4731	3000	0.3146	31.8340
0.2767	0.6309	4000	0.2984	31.7955
0.2908	0.7886	5000	0.2841	30.2510
0.2589	0.9463	6000	0.2702	28.5905
0.1941	1.1039	7000	0.2648	28.2867
0.1778	1.2617	8000	0.2613	27.8530
0.2072	1.4194	9000	0.2538	27.3528
0.1944	1.5771	10000	0.2466	26.9360
0.1889	1.7348	11000	0.2417	26.5371
0.1778	1.8925	12000	0.2361	25.7168
0.1384	2.0502	13000	0.2345	25.6473
0.1287	2.2079	14000	0.2345	25.8865
0.1335	2.3656	15000	0.2314	25.3250
0.1342	2.5233	16000	0.2286	24.9627
0.129	2.6810	17000	0.2255	24.7568
0.1213	2.8387	18000	0.2232	24.5562
0.1255	2.9965	19000	0.2215	24.5188

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu130
• Datasets 4.2.0
• Tokenizers 0.22.1