nandovallec/whisper-tiny-bg-l

Whisper Small Bg - Yonchevisky_tes2t

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7377
• Wer: 61.8352

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.8067	0.37	100	1.6916	137.6897
0.9737	0.73	200	1.1197	78.3571
0.7747	1.1	300	0.9763	73.8906
0.6672	1.47	400	0.8972	70.7102
0.6196	1.84	500	0.8329	67.4545
0.4849	2.21	600	0.7968	66.6029
0.4402	2.57	700	0.7597	62.7795
0.4601	2.94	800	0.7385	61.8642
0.3545	3.31	900	0.7394	61.5050
0.3596	3.68	1000	0.7377	61.8352

Framework versions

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