whisper-mmr-embeds-5h

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: 0.8001
• Wer: 28.6678

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: 5
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7309	0.3125	5	1.2018	37.5751
0.6763	0.625	10	1.1315	35.5722
0.6032	0.9375	15	1.0619	34.3409
0.5356	1.25	20	0.9966	34.8307
0.5222	1.5625	25	0.9379	34.5154
0.4771	1.875	30	0.8873	33.3646
0.4776	2.1875	35	0.8465	31.2913
0.4276	2.5	40	0.8170	29.4964
0.4784	2.8125	45	0.8001	28.6678

Framework versions

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