Gabi00/whisper-v3-LoRA-en_students_test

Whisper Small Eng - Gabriel Mora

This model is a fine-tuned version of openai/whisper-small on the English-mistakes dataset. It achieves the following results on the evaluation set:

• Loss: 0.3644
• Wer: 12.9853

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9139	0.1270	500	0.6388	24.1376
0.5572	0.2541	1000	0.4884	17.9087
0.5416	0.3811	1500	0.4371	15.2460
0.5542	0.5081	2000	0.4156	13.7921
0.6599	0.6352	2500	0.4036	13.4956
0.6117	0.7622	3000	0.3960	13.2676
0.5569	0.8892	3500	0.3890	13.1336
0.537	1.0163	4000	0.3850	12.5292
0.4677	1.1433	4500	0.3815	12.6261
0.5017	1.2703	5000	0.3792	12.4836
0.5346	1.3974	5500	0.3761	12.3126
0.4858	1.5244	6000	0.3735	12.2926
0.5478	1.6514	6500	0.3715	12.4009
0.5277	1.7785	7000	0.3699	12.2327
0.5153	1.9055	7500	0.3693	12.1643
0.5825	2.0325	8000	0.3681	12.1387
0.6049	2.1596	8500	0.3670	12.3211
0.5248	2.2866	9000	0.3662	12.1501
0.554	2.4136	9500	0.3653	12.0645
0.5031	2.5407	10000	0.3654	12.9312
0.5253	2.6677	10500	0.3647	12.9739
0.5132	2.7947	11000	0.3641	12.9511
0.5789	2.9217	11500	0.3644	12.9853

Framework versions

• PEFT 0.11.1
• Transformers 4.42.3
• Pytorch 2.1.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1