whisper-large-v3-turbo-lora-el

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1734
• Wer: 0.4643

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3662	10.4278	250	0.3702	0.5488
0.2983	20.8556	500	0.3048	0.5264
0.2742	31.2567	750	0.2839	0.5168
0.2647	41.6845	1000	0.2698	0.5096
0.242	52.0856	1250	0.2581	0.5085
0.2395	62.5134	1500	0.2471	0.5019
0.229	72.9412	1750	0.2371	0.4973
0.2187	83.3422	2000	0.2279	0.4955
0.2086	93.7701	2250	0.2192	0.4947
0.202	104.1711	2500	0.2113	0.4942
0.1952	114.5989	2750	0.2041	0.4936
0.1828	125.0	3000	0.1974	0.4805
0.1819	135.4278	3250	0.1918	0.4826
0.1748	145.8556	3500	0.1867	0.4786
0.1755	156.2567	3750	0.1825	0.4770
0.1719	166.6845	4000	0.1791	0.4708
0.169	177.0856	4250	0.1766	0.4707
0.1674	187.5134	4500	0.1748	0.4640
0.1662	197.9412	4750	0.1738	0.4643
0.1609	208.3422	5000	0.1734	0.4643

Framework versions

• PEFT 0.15.2
• Transformers 4.52.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1