whisper-20hrs-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: 0.3277
• Wer: 15.3521

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: 49
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6364	0.3025	49	0.5746	26.5240
0.4018	0.6049	98	0.3909	18.6802
0.3666	0.9074	147	0.3616	17.2711
0.3174	1.2099	196	0.3493	15.8956
0.3231	1.5123	245	0.3393	15.3957
0.2913	1.8148	294	0.3333	15.5131
0.2711	2.1173	343	0.3308	15.3655
0.2801	2.4198	392	0.3301	15.0166
0.278	2.7222	441	0.3277	15.3521

Framework versions

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