whisper-small-tigrinya-linear-r-32

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

• Cer: 15.6852
• Loss: 0.4011
• Wer: 43.8615

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 100
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
1.7104	0.6680	250	19.7292	0.4185	53.0521
1.2726	1.3340	500	16.7111	0.3549	46.3354
1.2303	2.0	750	15.7563	0.3362	44.6742
1.0308	2.6680	1000	15.2333	0.3258	43.4043
0.8122	3.3340	1250	15.7040	0.3314	43.8197
0.8717	4.0	1500	14.8800	0.3282	43.3715
0.7094	4.6680	1750	15.3272	0.3393	42.9562
0.5863	5.3340	2000	15.4644	0.3518	43.6434
0.6630	6.0	2250	15.3741	0.3435	43.0697
0.5734	6.6680	2500	15.3421	0.3620	43.5507
0.4406	7.3340	2750	15.5398	0.3816	43.8764
0.4993	8.0	3000	15.7726	0.3810	44.0975
0.4340	8.6680	3250	15.7936	0.3929	44.1872
0.3487	9.3340	3500	15.7022	0.4212	44.0258
0.4169	10.0	3750	15.6852	0.4011	43.8615

Framework versions

• PEFT 0.18.1
• Transformers 5.2.0
• Pytorch 2.9.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2