whisper-ssl-embeds-20h

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.3309
• Wer: 15.5266

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: 0.0001
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.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: 66
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5277	0.2973	66	0.4780	23.8870
0.388	0.5946	132	0.3847	17.5060
0.3642	0.8919	198	0.3622	17.3349
0.33	1.1892	264	0.3518	16.9188
0.3609	1.4865	330	0.3435	16.2110
0.3068	1.7838	396	0.3374	16.0365
0.3229	2.0811	462	0.3355	15.6742
0.3211	2.3784	528	0.3331	15.3018
0.3191	2.6757	594	0.3314	15.8419
0.316	2.9730	660	0.3309	15.5266

Framework versions

• PEFT 0.16.0
• Transformers 4.57.3
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.22.2