Ivan Lee

3 channels for mean, 3 channels for variance

import numpy as np from PIL import Image data = np.load('xxx.npz') k = 0 for i in data['arr_0']: if k%1000==0: print(k) img = Image.fromarray(i) img.save(str(k)+".jpg") k = k+1

Just use the "scripts/merge_lora_weights.py" to merge lora checkpoints into models before inference. Guidance:https://github.com/haotian-liu/LLaVA/blob/main/docs/LoRA.md

have you solved it?