whisper-maxcos-embeds-10h

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: 0.6154
• Wer: 22.1123

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: 7
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7226	0.2692	7	1.1895	37.2429
0.6575	0.5385	14	1.0908	34.1531
0.5641	0.8077	21	0.9930	36.8873
0.5019	1.0769	28	0.8992	34.2872
0.4553	1.3462	35	0.8104	26.7555
0.4343	1.6154	42	0.7343	25.8664
0.4374	1.8846	49	0.6846	24.8029
0.3686	2.1538	56	0.6528	23.6287
0.3535	2.4231	63	0.6312	22.6222
0.4244	2.6923	70	0.6193	22.1592
0.3497	2.9615	77	0.6154	22.1123

Framework versions

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