whisper-small-mn-8

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: 0.2421
• Wer: 26.5185
• Cer: 9.4681

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: 32
• eval_batch_size: 32
• 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: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3717	0.35	1000	0.4004	46.9576	16.9664
0.286	0.69	2000	0.3129	37.3935	13.5504
0.2287	1.04	3000	0.2768	33.1931	11.7806
0.2257	1.39	4000	0.2590	30.7243	11.0232
0.2029	1.73	5000	0.2428	29.2003	10.4144
0.1691	2.08	6000	0.2408	28.4357	10.0306
0.1626	2.43	7000	0.2369	28.0588	10.0486
0.1588	2.77	8000	0.2321	27.2340	9.6819
0.1271	3.12	9000	0.2349	26.8407	9.5574
0.1263	3.47	10000	0.2356	27.1630	9.6519
0.1314	3.81	11000	0.2340	26.5567	9.4278
0.1062	4.16	12000	0.2390	26.6332	9.5162
0.1081	4.5	13000	0.2398	26.5840	9.5085
0.1033	4.85	14000	0.2402	26.7096	9.4801
0.097	5.2	15000	0.2421	26.5185	9.4681

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2