whisper-mmr-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: 0.3946
• Wer: 16.2713

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: 21
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7733	0.2877	21	1.0613	34.8676
0.5207	0.5753	42	0.6361	23.4374
0.429	0.8630	63	0.4703	20.3107
0.3853	1.1507	84	0.4352	19.1968
0.3596	1.4384	105	0.4196	17.2879
0.3754	1.7260	126	0.4090	17.3650
0.3532	2.0137	147	0.4019	17.1470
0.3307	2.3014	168	0.3982	16.2579
0.34	2.5890	189	0.3958	16.4458
0.3289	2.8767	210	0.3946	16.2713

Framework versions

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