whisper-tiny-ga2en-v1.2

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. The best model (this version) based on ChrF is at checkpoint 900, epoch 13.24, and it achieves the following results on the evaluation set:

• Loss: 2.7626
• Bleu: 15.36
• Chrf: 32.05
• Wer: 93.1112

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Experiment

• learning_rate: 0.0001

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 128
• eval_batch_size: 128
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.03
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
0.8137	1.47	100	2.1667	8.99	24.09	125.4840
0.3542	2.94	200	2.2521	10.81	28.88	117.7398
0.0878	4.41	300	2.4337	14.69	31.31	98.6042
0.0602	5.88	400	2.5646	13.66	31.71	106.9338
0.0325	7.35	500	2.6156	13.97	30.21	99.7299
0.0323	8.82	600	2.6510	14.53	30.88	100.2701
0.0153	10.29	700	2.7360	14.17	30.99	103.1067
0.0092	11.76	800	2.7474	15.06	31.84	93.5615
0.0058	13.24	900	2.7626	15.36	32.05	93.1112
0.0039	14.71	1000	2.7774	15.38	32.04	93.6515

Framework versions

• Transformers 4.39.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2