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

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

• Loss: 0.1992
• Wer: 10.8477

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
0.1549	1.0	683	0.1534	17.4633
0.1252	2.0	1366	0.1311	14.6233
0.1128	3.0	2049	0.1281	14.2954
0.0943	4.0	2732	0.1195	13.4552
0.0849	5.0	3415	0.1189	13.1641
0.0788	6.0	4098	0.1184	12.7397
0.0687	7.0	4781	0.1170	12.5798
0.0628	8.0	5464	0.1189	12.5705
0.0553	9.0	6147	0.1189	11.9618
0.0473	10.0	6830	0.1207	12.0334
0.0396	11.0	7513	0.1224	11.9328
0.0332	12.0	8196	0.1312	11.7081
0.0254	13.0	8879	0.1343	11.6188
0.0187	14.0	9562	0.1417	11.5109
0.0133	15.0	10245	0.1528	11.3542
0.0089	16.0	10928	0.1637	11.2966
0.0057	17.0	11611	0.1724	11.1139
0.0034	18.0	12294	0.1837	11.0247
0.0020	19.0	12977	0.1918	10.8944
0.0011	20.0	13660	0.1992	10.8477

Framework versions

• PEFT 0.18.1
• Transformers 5.0.0
• Pytorch 2.9.0.dev20250821+rocm7.0.0.git125803b7
• Datasets 4.8.2
• Tokenizers 0.22.2