whisper-maxcos-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: 0.4605
• Wer: 17.9052

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: 16
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7063	0.2857	16	1.1193	35.2333
0.5128	0.5714	32	0.8445	27.9733
0.4019	0.8571	48	0.6201	22.1961
0.3671	1.1429	64	0.5204	20.2436
0.3611	1.4286	80	0.4896	20.3073
0.4283	1.7143	96	0.4779	18.1904
0.331	2.0	112	0.4702	19.1331
0.3912	2.2857	128	0.4658	18.1199
0.3416	2.5714	144	0.4629	18.5862
0.3787	2.8571	160	0.4605	17.9052

Framework versions

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