whisper-10hrs-meta-seed2025

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: 0.3596
• Wer: 15.9996

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: 24
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9264	0.2963	24	0.9643	34.3174
0.4812	0.5926	48	0.4858	19.8980
0.4231	0.8889	72	0.4138	20.4650
0.3738	1.1852	96	0.3907	18.2944
0.3299	1.4815	120	0.3781	18.2742
0.3514	1.7778	144	0.3686	17.3248
0.3199	2.0741	168	0.3639	16.2177
0.3573	2.3704	192	0.3614	16.7914
0.345	2.6667	216	0.3602	16.3049
0.3032	2.9630	240	0.3596	15.9996

Framework versions

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