whisper-40hrs-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: 0.3563
• Wer: 17.0161

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: 98
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2139	0.1529	50	1.2026	39.7759
1.1738	0.3058	100	1.0859	37.0182
1.0085	0.4587	150	0.8664	34.3007
0.8304	0.6116	200	0.6222	30.5130
0.6634	0.7645	250	0.4938	25.6181
0.6328	0.9174	300	0.4322	21.8137
0.5947	1.0703	350	0.4087	19.3646
0.5847	1.2232	400	0.3962	18.9553
0.6664	1.3761	450	0.3874	20.6998
0.5805	1.5291	500	0.3789	18.4889
0.5745	1.6820	550	0.3735	18.2776
0.5431	1.8349	600	0.3687	17.5328
0.5532	1.9878	650	0.3654	17.3416
0.556	2.1407	700	0.3627	17.2543
0.5581	2.2936	750	0.3607	17.1302
0.6142	2.4465	800	0.3588	17.0530
0.5406	2.5994	850	0.3574	17.0665
0.5222	2.7523	900	0.3567	17.0061
0.5753	2.9052	950	0.3563	17.0161

Framework versions

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