enakilci/whisper-large-v3-fi-800steps-8batch-4grad_steps-0.0001lr

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: 0.3318
• Wer: 25.9893

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: 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
1.3591	0.42	50	0.3653	32.0541
0.4623	0.84	100	0.4383	35.4316
0.3229	1.26	150	0.4386	35.3672
0.2538	1.68	200	0.4324	34.4929
0.1972	2.11	250	0.4287	34.4561
0.1194	2.53	300	0.4235	33.6094
0.1132	2.95	350	0.3826	30.1767
0.0669	3.37	400	0.4073	33.0941
0.0614	3.79	450	0.3869	29.5233
0.0435	4.21	500	0.3942	30.1859
0.032	4.63	550	0.3839	27.9404
0.0184	5.05	600	0.3571	25.7500
0.0094	5.47	650	0.3477	25.8605
0.0055	5.89	700	0.3371	26.6243
0.0026	6.32	750	0.3329	25.4463
0.0015	6.74	800	0.3318	25.9893

Framework versions

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