St4n/whisper-small-self

Whisper Small En - Stan

This model is a fine-tuned version of openai/whisper-small on the new-2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1269
• Wer: 8.5137

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0026	30.77	200	0.0885	3.0303
0.0001	61.54	400	0.1035	8.8023
0.0	92.31	600	0.1082	8.8023
0.0	123.08	800	0.1111	8.5137
0.0	153.85	1000	0.1128	8.5137
0.0	184.62	1200	0.1143	8.5137
0.0	215.38	1400	0.1153	8.5137
0.0	246.15	1600	0.1162	8.5137
0.0	276.92	1800	0.1169	8.5137
0.0	307.69	2000	0.1176	8.5137
0.0	338.46	2200	0.1196	8.5137
0.0	369.23	2400	0.1211	8.5137
0.0	400.0	2600	0.1217	8.5137
0.0	430.77	2800	0.1221	8.5137
0.0	461.54	3000	0.1224	8.5137
0.0	492.31	3200	0.1225	8.5137
0.0	523.08	3400	0.1227	8.5137
0.0	553.85	3600	0.1228	8.5137
0.0	584.62	3800	0.1229	8.5137
0.0	615.38	4000	0.1230	8.5137
0.0	646.15	4200	0.1253	8.5137
0.0	676.92	4400	0.1263	8.5137
0.0	707.69	4600	0.1265	8.5137
0.0	738.46	4800	0.1267	8.5137
0.0	769.23	5000	0.1266	8.5137
0.0	800.0	5200	0.1267	8.5137
0.0	830.77	5400	0.1267	8.5137
0.0	861.54	5600	0.1269	8.5137
0.0	892.31	5800	0.1269	8.5137
0.0	923.08	6000	0.1269	8.5137

Framework versions

• Transformers 4.39.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2