exp_002_base_lora

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

• Loss: 0.5781
• Wer: 54.5829
• Wer Ortho: 57.7728
• Cer: 17.2107

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: 32
• 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: 1000
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho	Cer
2.364	0.05	500	1.3775	89.3018	90.4210	30.2665
1.9822	0.1	1000	1.0647	79.1535	81.1369	25.7959
1.773	0.15	1500	0.8929	71.8646	74.4727	22.9622
1.6612	1.044	2000	0.8104	68.2874	70.8852	20.9447
1.5587	1.094	2500	0.7594	65.4784	68.2362	20.1626
1.5063	1.144	3000	0.7234	63.3287	66.1393	19.5334
1.4318	2.038	3500	0.6917	61.0404	64.1920	19.0537
1.3879	2.088	4000	0.6697	59.6591	62.7346	18.5726
1.3831	2.138	4500	0.6524	58.5128	61.6924	17.9122
1.3231	3.032	5000	0.6385	59.3316	62.5270	19.6542
1.3064	3.082	5500	0.6270	58.2777	61.2398	18.6335
1.2766	3.132	6000	0.6143	56.9425	60.0897	18.0292
1.255	4.026	6500	0.6063	57.1315	60.3056	18.3886
1.2481	4.076	7000	0.5986	56.2203	59.3298	17.9156
1.2161	4.126	7500	0.5905	55.6115	58.8274	17.8844
1.2109	5.02	8000	0.5860	55.6535	58.8233	17.7709
1.2183	5.07	8500	0.5834	55.3218	58.5202	17.4484
1.1953	5.12	9000	0.5781	54.5829	57.7728	17.2107
1.1721	6.014	9500	0.5760	54.8012	57.9347	17.4963
1.1816	6.064	10000	0.5753	54.7718	57.8641	17.3142

Framework versions

• PEFT 0.12.0
• Transformers 4.48.3
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4