library_name, license, base_model, tags, metrics, model-index, datasets

library_name	license	base_model	tags	metrics	model-index	datasets
transformers	mit	Sunbird/asr-whisper-large-v3-salt	generated_from_trainer	wer	name results cdli-whisper-ml-eng-lug-full-a40	cdli/ugandan_luganda_nonstandard_speech_v1.0 cdli/ugandan_english_nonstandard_speech_v1.0

cdli-whisper-ml-eng-lug-full-a40

This model is a fine-tuned version of Sunbird/asr-whisper-large-v3-salt on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7960
• Wer: 0.4932
• Cer: 0.3124
• test_cer = 0.1888
• test_loss = 0.5511
• test_runtime = 0:30:54.57
• test_samples_per_second = 1.1
• test_steps_per_second = 0.275
• test_wer = 0.3478

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: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 150
• training_steps: 2500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.7765	0.3615	250	0.8623	0.5111	0.3216
0.6632	0.7229	500	0.8226	0.5051	0.3236
0.5672	1.0839	750	0.8067	0.4862	0.3044
0.6058	1.4453	1000	0.7991	0.4949	0.3142
0.6589	1.8068	1250	0.7972	0.4901	0.3106
0.5959	2.1677	1500	0.7977	0.4926	0.3118
0.5402	2.5292	1750	0.7964	0.4926	0.3114
0.5934	2.8907	2000	0.7964	0.4921	0.3118
0.5464	3.2516	2250	0.7960	0.4931	0.3113
0.5497	3.6130	2500	0.7960	0.4932	0.3124

Framework versions

• Transformers 4.52.0
• Pytorch 2.7.1+cu118
• Datasets 3.6.0
• Tokenizers 0.21.4