l3-whisper-medium-l2c3_e4_v12

This model is a fine-tuned version of openai/whisper-medium on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2667
• Wer: 16.7248

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: 16
• eval_batch_size: 16
• 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: 500
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2269	0.2741	3000	0.2934	23.0264
0.1841	0.5482	6000	0.2579	20.6687
0.1617	0.8223	9000	0.2450	18.6915
0.1109	1.0964	12000	0.2430	17.9395
0.1104	1.3705	15000	0.2415	17.7986
0.1084	1.6446	18000	0.2366	17.5544
0.1047	1.9187	21000	0.2269	17.1679
0.0682	2.1928	24000	0.2472	16.7482
0.0686	2.4669	27000	0.2402	17.0820
0.0678	2.7410	30000	0.2434	17.0435
0.0579	3.0151	33000	0.2555	16.9403
0.0399	3.2892	36000	0.2642	16.7723
0.0392	3.5633	39000	0.2690	16.8861
0.0383	3.8374	42000	0.2667	16.7248

Framework versions

• Transformers 4.51.3
• Pytorch 2.8.0.dev20250319+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4