whisper-medium.en-decoder-layernorm-english-accent

This model is a fine-tuned version of openai/whisper-medium.en on the westbrook/English_Accent_DataSet dataset. It achieves the following results on the evaluation set:

• Loss: 1.1130
• Wer: 0.1202

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• 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: 0.05
• num_epochs: 2.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0	0	1.3238	0.1256
1.3097	0.1270	200	1.3095	0.1254
1.2627	0.2541	400	1.2871	0.1238
1.2548	0.3811	600	1.2666	0.1236
1.1832	0.5082	800	1.2468	0.1233
1.1767	0.6352	1000	1.2284	0.1228
1.2110	0.7623	1200	1.2111	0.1233
1.1878	0.8893	1400	1.1948	0.1235
1.1267	1.0159	1600	1.1796	0.1237
1.1965	1.1429	1800	1.1652	0.1236
1.0833	1.2700	2000	1.1519	0.1228
1.1211	1.3970	2200	1.1403	0.1225
1.0462	1.5241	2400	1.1303	0.1225
1.1257	1.6511	2600	1.1220	0.1204
1.1305	1.7781	2800	1.1163	0.1205
1.0701	1.9052	3000	1.1130	0.1202

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2