Spaces:

---

treble-technologies

/

ffasr

like 37

Running on CPU Upgrade

Hi FFASR maintainers,

Could you please review my pending submission when convenient?

Job ID: 107f6511 and c2b5cb5f
Model repo: banhchungtuongot/ffasr-pilot-x-adapter-weights
Base model: efficient-speech/lite-whisper-large-v3-turbo-acc
Base revision: ef2c0dd768cc9832a8a5a3397ab7218c838fea66

The submitted repository is public and contains only a 6.89 MB LoRA adapter.
The custom evaluator:

• downloads the public base model and public adapter from Hugging Face;
• attaches rank-16 LoRA to encoder self-attention and decoder cross-attention;
• does not access external services or private credentials;
• defines the required evaluate(file: pathlib.Path) -> str function;
• uses only torch, transformers, soundfile, safetensors, huggingface_hub and librosa.

I have locally verified that the adapter loads without missing or unexpected
LoRA keys and produces valid transcriptions.

Thank you for reviewing it.