whisper-maxcos-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: 0.9319
• Wer: 34.4248

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: 3
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7496	0.2308	3	1.2162	37.7898
0.7524	0.4615	6	1.1790	37.0819
0.6645	0.6923	9	1.1415	35.7366
0.6302	0.9231	12	1.1053	34.5523
0.5815	1.1538	15	1.0715	34.5758
0.5485	1.3846	18	1.0404	36.2801
0.5416	1.6154	21	1.0128	36.6223
0.5358	1.8462	24	0.9891	36.2801
0.5177	2.0769	27	0.9694	36.2331
0.547	2.3077	30	0.9538	34.1161
0.5061	2.5385	33	0.9423	34.2403
0.4961	2.7692	36	0.9349	35.1729
0.5108	3.0	39	0.9319	34.4248

Framework versions

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