whisper-a-nomi-18

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

• Loss: 0.0346
• Wer: 14.4772

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: 0.0004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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
• lr_scheduler_warmup_steps: 132
• num_epochs: 18
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.0	88	0.0813	11.4388
0.944	2.0	176	0.0636	11.2601
0.1726	3.0	264	0.0395	16.1752
0.0661	4.0	352	0.0895	25.7373
0.145	5.0	440	0.0627	19.9285
0.0218	6.0	528	0.0481	8.3110
0.0187	7.0	616	0.0782	23.0563
0.0282	8.0	704	0.0435	16.6220
0.0282	9.0	792	0.0284	11.7069
0.0055	10.0	880	0.0338	17.0688
0.0027	11.0	968	0.0463	17.3369
0.0039	12.0	1056	0.0362	11.6175
0.0038	13.0	1144	0.0353	14.6559
0.0014	14.0	1232	0.0347	14.5666
0.0	15.0	1320	0.0346	14.4772
0.0	16.0	1408	0.0346	14.4772
0.0	17.0	1496	0.0346	14.4772
0.0	18.0	1584	0.0346	14.4772

Framework versions

• Transformers 4.48.0.dev0
• Pytorch 2.4.0
• Datasets 3.1.0
• Tokenizers 0.21.0