ft-C-convert-for-SDXL-comfyUI-longCLIP.py

import torch
import os

# Load the fine-tuned model checkpoint of your choice
device = 'cuda'
THE_FINETUNED_MODEL = torch.load('ft-checkpoints/longclip_ft_10.pt', map_location=device)

# Save only the state dictionary of the fine-tuned model
# This can be used with SDXL ComfyUI, see -> https://github.com/SeaArtLab/ComfyUI-Long-CLIP
torch.save(THE_FINETUNED_MODEL.state_dict(), 'ft-checkpoints/longclip_ft_10_STATE_DICT.pt')