nurzhanit/whisper-omg-2.5.2

Whisper Small Hi - Sanchit Gandhi

This model is a fine-tuned version of nurzhanit/whisper-enhanced-ml on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0018
• Wer: 0.0939

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 25
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1382	0.7246	50	0.0746	12.5396
0.0509	1.4493	100	0.0487	6.8804
0.0476	2.1739	150	0.0270	4.9548
0.0328	2.8986	200	0.0165	3.1114
0.0167	3.6232	250	0.0098	1.7494
0.0078	4.3478	300	0.0061	0.8454
0.0056	5.0725	350	0.0040	0.3640
0.0033	5.7971	400	0.0026	0.2231
0.0024	6.5217	450	0.0020	0.1409
0.0021	7.2464	500	0.0018	0.0939

Framework versions

• Transformers 4.40.0
• Pytorch 2.5.0+cu124
• Datasets 3.0.2
• Tokenizers 0.19.1