qmeeus/whisper-small-ner-combined

WhisperForNamedEntityRecognition

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

• Loss: 8.1514
• Wer: 10.4828

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: 0.0001
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 1600
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
31.3741	0.06	100	25.8582	10.4828
13.0078	1.03	200	13.4173	10.4828
10.3619	1.09	300	10.8540	10.4828
8.7869	2.06	400	9.6249	10.4828
7.3964	3.02	500	9.1812	10.4828
6.6321	3.08	600	8.6536	10.4828
6.4612	4.05	700	8.6046	10.4828
4.8358	5.02	800	8.0890	10.4828
4.4918	5.08	900	8.3141	10.4828
4.7548	6.04	1000	8.1660	10.4828
3.7881	7.01	1100	8.2471	10.4828
3.1916	7.07	1200	8.0779	10.4828
3.2039	8.04	1300	8.1106	10.4828
3.038	9.0	1400	8.0875	10.4828
2.3249	9.07	1500	8.1025	10.4828
2.6124	10.03	1600	8.1514	10.4828

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.10.0
• Datasets 2.7.1.dev0
• Tokenizers 0.11.0