psst-model-4e-1s-augmented

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

• Loss: 0.5793
• Wer: 0.1566
• Iu F1: 0.7215
• Iu Tp: 829
• Iu Fp: 471
• Iu Fn: 169

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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: 1327
• training_steps: 18952

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Iu F1	Iu Tp	Iu Fp	Iu Fn
0.5869	0.5001	2369	0.5099	0.2429	0.7513	420	122	156
0.2396	1.0	4738	0.5600	0.2424	0.7090	452	247	124
0.0832	1.5001	7107	0.6434	0.2410	0.7457	409	112	167
0.0351	2.0	9476	0.7193	0.2330	0.7571	438	143	138
0.0151	2.5001	11845	0.7708	0.2348	0.7006	468	292	108
0.0064	3.0	14214	0.8060	0.2297	0.7753	459	149	117
0.0022	3.5001	16583	0.8301	0.2237	0.7787	445	122	131
0.0012	4.0	18952	0.8656	0.2243	0.7699	445	135	131

Framework versions

• Transformers 5.6.2
• Pytorch 2.6.0+cu124
• Datasets 2.21.0
• Tokenizers 0.22.2