exp4-whisper-all-jasmin

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: 0.3347
• Wer: 16.0466

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: 1e-05
• 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: 150
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9942	0.2	100	1.1448	36.9041
0.7153	0.4	200	0.7109	33.2237
0.4752	0.6	300	0.4312	21.6157
0.4388	0.8	400	0.3911	20.1932
0.4234	1.0	500	0.3737	18.9419
0.417	1.2	600	0.3623	18.1199
0.4012	1.4	700	0.3544	17.8549
0.3898	1.6	800	0.3487	17.5127
0.4018	1.8	900	0.3445	17.3785
0.3736	2.0	1000	0.3415	15.7815
0.3804	2.2	1100	0.3389	16.2277
0.397	2.4	1200	0.3369	16.1338
0.3772	2.6	1300	0.3356	16.0935
0.3781	2.8	1400	0.3350	16.5968
0.3675	3.0	1500	0.3347	16.0466

Framework versions

• PEFT 0.16.0
• Transformers 4.52.0
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2