enakilci/whisper-large-v3-fi-800steps-8batch-8grad_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: 0.3858
• Wer: 21.6363

It achieves the following results on the Test set:

• Eval_Wer: 21.636296705319342
• Eval_NormalizedWer: 18.727590328215502

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: 8
• total_train_batch_size: 64
• 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
0.0049	0.84	50	0.4045	27.8851
0.0264	1.68	100	0.4482	29.3852
0.0189	2.53	150	0.4076	26.6980
0.0129	3.37	200	0.3772	24.5905
0.0087	4.21	250	0.3875	25.5108
0.0054	5.05	300	0.3754	24.9034
0.0035	5.89	350	0.3742	23.5505
0.0014	6.74	400	0.3823	23.4677
0.0014	7.58	450	0.3914	23.5781
0.0012	8.42	500	0.3771	22.3173
0.0007	9.26	550	0.3812	21.8756
0.0002	10.11	600	0.3812	21.7191
0.0002	10.95	650	0.3825	21.6547
0.0001	11.79	700	0.3844	21.6363
0.0001	12.63	750	0.3854	21.5995
0.0001	13.47	800	0.3858	21.6363

Framework versions

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