SPEAK-ASR/whisper-si-exp-10-medium

This model is a fine-tuned version of openai/whisper-medium on the SPEAK-ASR/openslr-sinhala-asr-norm-noise-rem-preprocessed dataset. It achieves the following results on the evaluation set:

• Loss: 0.1862
• Wer: 9.9913

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: 0.0006926233815806411
• train_batch_size: 128
• eval_batch_size: 256
• seed: 42
• optimizer: Use 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: 295
• num_epochs: 20.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1513	1.0	657	0.1421	15.8280
0.1196	2.0	1314	0.1214	13.7379
0.1062	3.0	1971	0.1152	12.8573
0.0945	4.0	2628	0.1139	12.6160
0.0830	5.0	3285	0.1504	15.6556
0.0755	6.0	3942	0.1068	11.4741
0.0667	7.0	4599	0.1063	11.5538
0.0604	8.0	5256	0.1083	11.3148
0.0526	9.0	5913	0.1083	11.1702
0.0464	10.0	6570	0.1108	11.0489
0.0388	11.0	7227	0.1133	10.9288
0.0315	12.0	7884	0.1204	10.7769
0.0243	13.0	8541	0.1236	10.7757
0.0188	14.0	9198	0.1314	10.7273
0.0127	15.0	9855	0.1409	10.4382
0.0089	16.0	10512	0.1494	10.4137
0.0054	17.0	11169	0.1604	10.2128
0.0032	18.0	11826	0.1710	10.1314
0.0019	19.0	12483	0.1803	10.0693
0.0012	20.0	13140	0.1862	9.9913

Framework versions

• PEFT 0.18.1
• Transformers 5.0.0
• Pytorch 2.9.0.dev20250821+rocm7.0.0.git125803b7
• Datasets 4.5.0
• Tokenizers 0.22.2