Noise issue after LoRA merge and inference

[positivewon]

❌ Actual Behavior

Hello, thank you very much for open-sourcing the OpenAudio code.

I’m currently fine-tuning the model with LoRA, but after merging the LoRA weights and running inference using the merged checkpoint, the output contains unexpected noise.

Could you please let me know what might be causing this issue, or if there’s a recommended way to properly merge and use the LoRA weights for inference?

Thank you for your time and support.

[zhang-heng]

same here.