whisper-20hrs-meta

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: 0.4035
• Wer: 19.7571

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: 55
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0357	0.1374	25	1.2179	38.0448
1.0051	0.2747	50	1.1769	37.1523
0.9812	0.4121	75	1.0983	36.4377
0.8721	0.5495	100	1.0066	35.1360
0.8113	0.6868	125	0.9084	34.5087
0.7209	0.8242	150	0.8029	32.2676
0.6349	0.9615	175	0.6938	31.7241
0.5759	1.0989	200	0.6019	29.6105
0.5171	1.2363	225	0.5404	26.2220
0.4895	1.3736	250	0.4981	25.0679
0.458	1.5110	275	0.4668	23.3066
0.4482	1.6484	300	0.4468	22.2230
0.4274	1.7857	325	0.4336	21.5620
0.4331	1.9231	350	0.4246	20.5455
0.418	2.0604	375	0.4185	20.5120
0.4347	2.1978	400	0.4140	20.0456
0.4128	2.3352	425	0.4101	19.9014
0.403	2.4725	450	0.4074	19.1834
0.4331	2.6099	475	0.4055	19.1230
0.3932	2.7473	500	0.4043	19.7370
0.4123	2.8846	525	0.4035	19.7571

Framework versions

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