b1n1yam/shook-tiny-amharic-stage2-polish

tiny Amharic - Biniyam Daniel

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0627
• Wer: 20.7634

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: 128
• eval_batch_size: 128
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 256
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0595	0.2344	300	0.0684	22.4376
0.0582	0.4688	600	0.0675	22.1963
0.0591	0.7031	900	0.0668	22.0622
0.0595	0.9375	1200	0.0657	21.6653
0.0551	1.1719	1500	0.0651	21.5516
0.0559	1.4062	1800	0.0644	21.2442
0.055	1.6406	2100	0.0639	21.1184
0.0547	1.875	2400	0.0634	20.9395
0.0519	2.1094	2700	0.0632	20.8622
0.0526	2.3438	3000	0.0630	20.7644
0.051	2.5781	3300	0.0628	20.7634
0.0511	2.8125	3600	0.0627	20.7634

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1
• Datasets 3.6.0
• Tokenizers 0.20.3