enakilci/whisper-large-v3-fi-800steps-8batch-4grad_steps-5e-05lr

Whisper Large v3 Fine-Tuned Finnish - CommonVoice13

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2383
• Wer: 24.6733

It achieves the following results on the test set:

• Wer: 17.043990428860667
• normalized_wer: 14.149122000551623

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: 50
• training_steps: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5438	0.42	50	0.2371	23.3204
0.2701	0.84	100	0.2817	26.4035
0.1849	1.26	150	0.2757	26.3851
0.1282	1.68	200	0.2701	23.1640
0.1066	2.11	250	0.2573	22.6210
0.0611	2.53	300	0.2782	23.8082
0.062	2.95	350	0.2611	23.0167
0.033	3.37	400	0.2760	23.5321
0.0315	3.79	450	0.2745	25.0874
0.0194	4.21	500	0.2567	20.3111
0.0119	4.63	550	0.2399	24.1119
0.0085	5.05	600	0.2379	22.3541
0.0026	5.47	650	0.2417	21.5995
0.0025	5.89	700	0.2366	22.1701
0.0009	6.32	750	0.2370	24.8297
0.0006	6.74	800	0.2383	24.6733

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.0.1
• Datasets 2.16.1
• Tokenizers 0.15.0