whisper-large-no-is-145h-30k-steps

This model is a fine-tuned version of NbAiLab/nb-whisper-large on the samromur_asr dataset. It achieves the following results on the evaluation set:

• Loss: 0.1280
• Wer: 8.0203

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3237	0.1778	1000	0.3331	24.9280
0.2334	0.3556	2000	0.2369	19.2033
0.1932	0.5333	3000	0.2121	16.7584
0.1681	0.7111	4000	0.1874	14.9669
0.1487	0.8889	5000	0.1720	13.8274
0.0911	1.0667	6000	0.1644	13.2541
0.0765	1.2444	7000	0.1577	12.3332
0.0843	1.4222	8000	0.1469	11.8722
0.0797	1.6	9000	0.1437	11.3156
0.0778	1.7778	10000	0.1365	10.9298
0.0789	1.9556	11000	0.1329	10.7554
0.039	2.1333	12000	0.1353	10.3625
0.0394	2.3111	13000	0.1399	10.2872
0.0438	2.4889	14000	0.1357	10.0926
0.0365	2.6667	15000	0.1287	9.8632
0.0406	2.8444	16000	0.1266	9.6435
0.0193	3.0222	17000	0.1284	9.3258
0.0181	3.2	18000	0.1309	9.3019
0.0215	3.3778	19000	0.1317	9.1299
0.0207	3.5556	20000	0.1287	8.8767
0.0185	3.7333	21000	0.1285	8.8731
0.0162	3.9111	22000	0.1272	8.6581
0.007	4.0889	23000	0.1285	8.6175
0.0076	4.2667	24000	0.1296	8.3655
0.0074	4.4444	25000	0.1277	8.4037
0.0076	4.6222	26000	0.1289	8.3464
0.006	4.8	27000	0.1257	8.1756
0.0067	4.9778	28000	0.1252	8.0979
0.0025	5.1556	29000	0.1269	8.0322
0.0032	5.3333	30000	0.1280	8.0203

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1