whisper-large-v3-med-pl-lora-decoder-only

This model is a fine-tuned version of openai/whisper-large-v3 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2346
• Model Preparation Time: 0.031
• Wer: 19.6622
• Cer: 5.5539

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: 16
• eval_batch_size: 8
• 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_ratio: 0.1
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
0.2073	0.4990	250	0.2922	0.031	23.1959	6.5656
0.166	0.9980	500	0.2567	0.031	21.3155	5.8382
0.1681	1.4970	750	0.2395	0.031	20.4712	5.5663
0.1604	1.9960	1000	0.2308	0.031	20.2512	5.6823
0.1402	2.4950	1250	0.2248	0.031	20.1093	5.6366
0.1604	2.9940	1500	0.2208	0.031	20.6273	6.0950
0.1308	3.4930	1750	0.2205	0.031	19.0023	5.1251
0.1307	3.9920	2000	0.2172	0.031	19.0520	5.1156
0.125	4.4910	2250	0.2216	0.031	21.5071	7.3168
0.1266	4.9900	2500	0.2168	0.031	19.0733	5.1070
0.1037	5.4890	2750	0.2203	0.031	19.1655	5.1032
0.1253	5.9880	3000	0.2179	0.031	18.9243	5.2050
0.1095	6.4870	3250	0.2232	0.031	20.0028	6.2338
0.1049	6.9860	3500	0.2226	0.031	18.6121	5.0015
0.0962	7.4850	3750	0.2275	0.031	19.2365	5.1289
0.0967	7.9840	4000	0.2259	0.031	19.1017	5.0785
0.0929	8.4830	4250	0.2307	0.031	19.3855	5.1146
0.0959	8.9820	4500	0.2294	0.031	18.9385	5.0766
0.0857	9.4810	4750	0.2361	0.031	20.1093	5.9076
0.0945	9.9800	5000	0.2346	0.031	19.6622	5.5539

Framework versions

• PEFT 0.17.1
• Transformers 4.51.3
• Pytorch 2.8.0+cu128
• Datasets 4.3.0
• Tokenizers 0.21.4