enakilci/whisper-large-v3-fi-1600steps-8batch-2grad_steps-0.0001lr

Whisper Large v3 Fine-Tuned Finnish

This model is a fine-tuned version of openai/whisper-large-v3 on the mozilla-foundation/common_voice_13_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3970
• Wer: 27.1397

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: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 1600
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7157	0.21	50	0.4892	42.8216
0.6602	0.42	100	0.6444	58.3379
0.5952	0.63	150	0.6134	52.3560
0.5649	0.84	200	0.5645	53.3499
0.5078	1.05	250	0.5867	69.4736
0.289	1.26	300	0.6150	55.5034
0.3318	1.47	350	0.5309	40.3000
0.2859	1.68	400	0.5462	41.3584
0.3091	1.89	450	0.4891	38.6159
0.2089	2.11	500	0.5305	41.0915
0.148	2.32	550	0.5124	37.8888
0.1538	2.53	600	0.5075	36.9225
0.1515	2.74	650	0.5187	37.9440
0.1532	2.95	700	0.4666	35.4040
0.0982	3.16	750	0.4934	36.7385
0.0832	3.37	800	0.4796	34.3641
0.0782	3.58	850	0.4742	39.5270
0.0724	3.79	900	0.4634	42.7388
0.0733	4.0	950	0.4544	38.6987
0.0409	4.21	1000	0.4778	33.1953
0.0437	4.42	1050	0.4584	33.3793
0.036	4.63	1100	0.4476	33.9223
0.0321	4.84	1150	0.4538	32.0725
0.0257	5.05	1200	0.4390	37.2538
0.0133	5.26	1250	0.4319	30.7289
0.0104	5.47	1300	0.4065	30.5264
0.0122	5.68	1350	0.4012	28.5754
0.0071	5.89	1400	0.4113	28.4925
0.0038	6.11	1450	0.3939	26.8820
0.0019	6.32	1500	0.3960	26.8820
0.0021	6.53	1550	0.3978	27.1305
0.0017	6.74	1600	0.3970	27.1397

Framework versions

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