whisper-finetuned-v3_3kstep

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

• Loss: 0.0553
• Wer: 51.7226

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: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• 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
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6724	0.1119	100	1.2435	253.1096
1.0385	0.2237	200	0.7294	550.1566
0.4208	0.3356	300	0.2787	107.8300
0.2423	0.4474	400	0.2023	74.1834
0.2164	0.5593	500	0.1818	86.8009
0.1902	0.6711	600	0.1494	72.5280
0.1487	0.7830	700	0.1384	77.0022
0.1426	0.8949	800	0.1195	69.1723
0.1182	1.0067	900	0.1125	67.5615
0.0937	1.1186	1000	0.1082	66.9351
0.0944	1.2304	1100	0.1022	64.2506
0.0875	1.3423	1200	0.0987	62.5503
0.0846	1.4541	1300	0.0869	62.0582
0.0815	1.5660	1400	0.0876	61.2528
0.0731	1.6779	1500	0.0815	59.9105
0.0703	1.7897	1600	0.0739	57.5839
0.0681	1.9016	1700	0.0707	57.5839
0.0618	2.0134	1800	0.0733	57.4049
0.0368	2.1253	1900	0.0680	57.1812
0.0417	2.2371	2000	0.0674	57.6734
0.0346	2.3490	2100	0.0656	55.8837
0.0416	2.4609	2200	0.0622	56.0179
0.0348	2.5727	2300	0.0625	56.1074
0.0338	2.6846	2400	0.0599	56.7338
0.0351	2.7964	2500	0.0565	53.6913
0.031	2.9083	2600	0.0561	55.1678
0.0262	3.0201	2700	0.0559	51.8568
0.0158	3.1320	2800	0.0555	52.0358
0.015	3.2438	2900	0.0561	51.9016
0.0179	3.3557	3000	0.0553	51.7226

Framework versions

• Transformers 4.51.3
• Pytorch 2.7.0+cu126
• Datasets 3.5.1
• Tokenizers 0.21.1