whisper-large-v3-el_tedx

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

• Loss: 1.8323
• Wer: 0.8544

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
1.842	31.2807	250	1.9497	0.9275
1.5801	62.5614	500	1.7260	0.9010
1.5127	93.8421	750	1.6892	0.9010
1.4104	125.0	1000	1.6756	0.9017
1.4227	156.2807	1250	1.6738	0.8962
1.3884	187.5614	1500	1.6796	0.8746
1.3471	218.8421	1750	1.6912	0.8578
1.2624	250.0	2000	1.7049	0.8564
1.2821	281.2807	2250	1.7208	0.8606
1.2589	312.5614	2500	1.7367	0.8697
1.2349	343.8421	2750	1.7520	0.8578
1.1661	375.0	3000	1.7658	0.8551
1.1958	406.2807	3250	1.7797	0.8530
1.1841	437.5614	3500	1.7933	0.8544
1.1712	468.8421	3750	1.8055	0.8787
1.1095	500.0	4000	1.8162	0.8794
1.1482	531.2807	4250	1.8238	0.8557
1.154	562.5614	4500	1.8279	0.8530
1.142	593.8421	4750	1.8325	0.8544
1.0939	625.0	5000	1.8323	0.8544

Framework versions

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