fs-w-he-tiny

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.6276
• Wer: 139.9335
• Cer: 134.1807

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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
10.7026	4.5872	500	10.1826	184.1406	154.3198
2.9461	9.1743	1000	3.4468	227.1130	234.3696
1.2202	13.7615	1500	2.2670	130.1045	120.0275
0.5577	18.3486	2000	1.9164	113.2953	101.0735
0.2603	22.9358	2500	1.9956	138.4615	129.8952
0.0533	27.5229	3000	2.1263	130.7217	120.5428
0.012	32.1101	3500	2.3153	111.6334	96.2985
0.0012	36.6972	4000	2.4891	126.2108	114.8574
0.0003	41.2844	4500	2.5962	139.4112	129.4401
0.0002	45.8716	5000	2.6276	139.9335	134.1807

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0