LoRA loader for nunchaku transformer
Thanks for releasing the Nunchaku quantized version. I’m just wondering how to load an unquantized LoRA into this quantized Nunchaku transformer?
About lora
LoRA support for FLUX.2-klein-9B-Nunchaku has been released.Please copy the files from the repository to the corresponding directories:
nunchaku/lora/common/
nunchaku/models/transformers/transformer_flux2.py
nunchaku/torch_transfer_utils.py
Usage
lora_path = "/home/tonera/loras/ultra_real_v2.safetensors"
transformer.update_lora_params(lora_path)
transformer.set_lora_strength(0.8)
BTW: How to locate the Nunchaku installation path and copy the above files to the correct directories? Run the following command:
python -c "import nunchaku; print(nunchaku.file)"
You will get output similar to this:/home/tonera/miniconda3/envs/aiservice/lib/python3.11/site-packages/nunchaku/init.py Here, /home/tonera/miniconda3/envs/aiservice/lib/python3.11/site-packages/nunchaku is your Nunchaku installation directory. Simply copy the aforementioned files into the matching subdirectories under this path.