VMadalina/whisper-small_ron3ws-music

Whisper Small Ro - VM6

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.2294
• Wer: 46.5315
• Cer: 19.8241

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8374	0.92	250	1.0621	59.3526	30.1464
0.4817	1.85	500	0.9239	50.9594	22.8332
0.186	2.77	750	0.9620	49.2374	21.9476
0.0634	3.69	1000	1.0351	49.5621	21.8295
0.0231	4.61	1250	1.1032	50.3395	22.4809
0.0106	5.54	1500	1.1072	55.5545	25.3680
0.0077	6.46	1750	1.1589	47.1908	20.5936
0.0036	7.38	2000	1.1499	49.2669	21.8019
0.0057	8.3	2250	1.1656	47.4073	20.5404
0.0018	9.23	2500	1.1767	47.1810	20.5621
0.0014	10.15	2750	1.1721	46.8562	20.2413
0.0015	11.07	3000	1.1905	46.6890	20.0405
0.0009	11.99	3250	1.1961	46.4528	19.9559
0.0006	12.92	3500	1.2025	47.2597	20.3849
0.0006	13.84	3750	1.2141	46.6103	20.0287
0.0008	14.76	4000	1.2178	46.4233	19.9658
0.0005	15.68	4250	1.2231	46.3249	19.8457
0.0004	16.61	4500	1.2265	46.6004	19.8910
0.0003	17.53	4750	1.2288	46.5512	19.8201
0.0003	18.45	5000	1.2294	46.5315	19.8241

Framework versions

• Transformers 4.32.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3