whisper-ssl-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: 0.3849
• Wer: 19.4115

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: 23
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3466	0.2911	23	0.9769	35.7936
0.5737	0.5823	46	0.5062	23.4744
0.4855	0.8734	69	0.4380	21.5956
0.3697	1.1646	92	0.4151	20.0154
0.4426	1.4557	115	0.4030	18.6600
0.3947	1.7468	138	0.3952	18.9721
0.4336	2.0380	161	0.3906	19.3411
0.3908	2.3291	184	0.3869	19.0962
0.41	2.6203	207	0.3861	19.3881
0.3571	2.9114	230	0.3849	19.4115

Framework versions

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