geninhu/whisper-large-v2-multiset-vi

openai/whisper-large-v2

This model is a fine-tuned version of openai/whisper-large-v2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4041
• Wer: 15.7710
• Cer: 7.6691

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Training data:

• mozilla-foundation/common_voice_11_0
• google/fleurs

Evaluation data:

• mozilla-foundation/common_voice_11_0

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-07
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 3
• total_train_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3983	0.1	500	0.5338	19.5876	10.6391
0.2277	1.08	1000	0.4134	16.5826	8.2668
0.172	2.05	1500	0.3968	16.3084	7.9787
0.1823	3.03	2000	0.3956	16.1768	7.8159
0.1445	4.0	2500	0.3955	16.0342	7.7438
0.147	4.1	3000	0.3965	15.8807	7.7145
0.1292	5.08	3500	0.4000	15.8587	7.7065
0.1187	6.05	4000	0.4029	15.7491	7.6398
0.1368	7.03	4500	0.4041	15.7600	7.6558
0.1231	8.0	5000	0.4041	15.7710	7.6691

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2