whisper-large-v3-natbed-combined-model

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: 0.7208
• Wer: 51.8794

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: 1.75e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.545	0.3125	250	0.8722	68.7694
0.8497	0.625	500	0.7933	60.7207
0.7736	0.9375	750	0.7221	58.3919
0.6343	1.25	1000	0.7073	56.5615
0.6016	1.5625	1250	0.6753	52.1000
0.5695	1.875	1500	0.6599	53.8078
0.4779	2.1875	1750	0.6836	51.0378
0.4058	2.5	2000	0.6781	49.7303
0.4294	2.8125	2250	0.6692	51.9284
0.36	3.125	2500	0.7229	51.1276
0.2672	3.4375	2750	0.7208	51.8794

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1