Configuration Parsing Warning:In adapter_config.json: "peft.task_type" must be a string
Whisper Small Sinhala LoRA (rank 32, seed 456)
Fresh rank-32 PEFT adapter trained on Sinhala speech with seed 456.
The frozen Whisper base weights are not included. LoRA targets q_proj and
v_proj in encoder self-attention, decoder self-attention, and decoder
cross-attention.
- Base model:
openai/whisper-small - Training rows:
32062 - Validation rows:
4171 - Held-out test rows:
4326 - Test WER (ratio):
0.4858421000884869 - Test CER (ratio):
0.1259456321111968
- Downloads last month
- 24
Model tree for dehanns/whisper-small-sinhala-lora-r32-seed-456
Base model
openai/whisper-small