enakilci/whisper-large-v3-fi-800steps-8batch-8grad_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: 0.3365
• Wer: 20.4123

It achieves the following results on the Test set:

• Eval_Wer: 20.430701270016566
• Eval_NormalizedWer: 17.3945021605222

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: 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
1.4113	0.84	50	0.3499	29.6797
0.2435	1.68	100	0.3457	29.3392
0.1657	2.53	150	0.3645	28.0692
0.0981	3.37	200	0.3741	28.5478
0.0651	4.21	250	0.4005	29.8454
0.0471	5.05	300	0.3751	28.0968
0.0291	5.89	350	0.3521	25.7869
0.0178	6.74	400	0.3535	24.1303
0.0099	7.58	450	0.3393	23.2468
0.0053	8.42	500	0.3336	23.6702
0.0031	9.26	550	0.3284	21.9676
0.0018	10.11	600	0.3328	22.9615
0.0008	10.95	650	0.3324	20.6976
0.0003	11.79	700	0.3348	20.5227
0.0003	12.63	750	0.3360	20.4215
0.0002	13.47	800	0.3365	20.4123

Framework versions

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