whisper-ssl-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: 0.3550
• Wer: 16.6035

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: 41
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8358	0.2950	41	0.6547	31.7912
0.4774	0.5899	82	0.4206	21.1595
0.456	0.8849	123	0.3900	18.8311
0.3977	1.1799	164	0.3766	18.4017
0.3519	1.4748	205	0.3690	18.1669
0.3475	1.7698	246	0.3624	16.6672
0.3545	2.0647	287	0.3591	16.7410
0.3348	2.3597	328	0.3576	16.7947
0.343	2.6547	369	0.3559	16.5934
0.3634	2.9496	410	0.3550	16.6035

Framework versions

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