whisper-tiny-no-specific-topic-v4

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

• Loss: 0.9218
• Wer: 41.2727

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0787	0.05	400	1.1607	69.0727
0.6467	0.1	800	0.7705	46.6545
0.4972	0.15	1200	0.7713	47.2727
0.3384	0.2	1600	0.7776	53.0455
0.3232	0.25	2000	0.8289	49.5545
0.2959	0.3	2400	0.8160	41.5
0.3069	0.35	2800	0.8336	41.8818
0.2521	0.4	3200	0.8755	43.4364
0.2477	0.45	3600	0.9011	42.5545
0.2178	0.5	4000	0.8802	42.1091
0.2579	0.55	4400	0.9062	44.5818
0.2687	0.6	4800	0.9092	41.8182
0.2393	0.65	5200	0.9255	43.0091
0.2682	0.7	5600	0.9155	43.0273
0.1842	0.75	6000	0.9215	42.8909
0.2081	0.8	6400	0.9269	43.8
0.2619	0.85	6800	0.9183	41.8091
0.2679	0.9	7200	0.9196	41.5182
0.2172	0.95	7600	0.9220	41.2455
0.1914	1.0	8000	0.9218	41.2727

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.2
• Tokenizers 0.21.0