enakilci/whisper-large-v3-fi-800steps-16batch-2grad_steps-0.0001lr

Whisper Large v3 Fine-Tuned Finnish

This model is a fine-tuned version of openai/whisper-large-v3 on the Common Voice 13.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3344
• Wer: 23.0167

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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
0.5817	0.42	50	0.4090	37.5023
0.4669	0.84	100	0.4374	35.8274
0.3154	1.26	150	0.4848	39.0484
0.2192	1.68	200	0.4313	34.6954
0.1985	2.1	250	0.4346	34.5205
0.1125	2.52	300	0.4307	32.8640
0.1039	2.94	350	0.4278	31.3271
0.067	3.36	400	0.4043	33.5542
0.0577	3.78	450	0.3911	40.7050
0.0461	4.2	500	0.3966	30.4712
0.0264	4.62	550	0.3630	27.2041
0.0204	5.04	600	0.3632	26.0353
0.0092	5.46	650	0.3448	24.4156
0.006	5.88	700	0.3284	23.9278
0.002	6.3	750	0.3334	23.2836
0.0019	6.72	800	0.3344	23.0167

Framework versions

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