whisper_large_v3_turbo_noise_redux_v2

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

• Loss: 6.8827

Model description

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 6
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss
11.0793	1.0	12	10.8135
10.6542	2.0	24	10.5037
10.4376	3.0	36	10.3954
10.2174	4.0	48	9.9744
9.7778	5.0	60	9.5521
9.3734	6.0	72	9.1692
9.0088	7.0	84	8.8258
8.6836	8.0	96	8.5225
8.399	9.0	108	8.2601
8.1554	10.0	120	8.0387
7.9527	11.0	132	7.8584
7.7913	12.0	144	7.7191
7.6702	13.0	156	7.6199
7.5894	14.0	168	7.5602
7.5475	15.0	180	7.5394
7.5068	16.0	192	7.4662
7.4412	17.0	204	7.4174
7.407	18.0	216	7.4003
7.3431	19.0	228	7.2668
7.2094	20.0	240	7.1460
7.0999	21.0	252	7.0497
7.0143	22.0	264	6.9762
6.9508	23.0	276	6.9245
6.9086	24.0	288	6.8935
6.8869	25.0	300	6.8827

Framework versions

• Transformers 4.39.3
• Pytorch 2.4.1+cu124
• Datasets 2.18.0
• Tokenizers 0.15.2