mg.py

import os
import torch
from safetensors.torch import load_file, save_file  # Updated import

def merge_models(model_path_1, model_path_2, ratio_1, ratio_2, output_path):
    # Ensure the sum of ratios equals 1
    total_ratio = ratio_1 + ratio_2
    if total_ratio != 1:
        ratio_1 /= total_ratio
        ratio_2 /= total_ratio

    print(f"Merging models with ratios: {ratio_1:.2f} (Model 1) and {ratio_2:.2f} (Model 2)")

    # Check if model paths exist
    if not (os.path.exists(model_path_1) and os.path.exists(model_path_2)):
        raise FileNotFoundError("One or both model files do not exist.")

    try:
        # Load the models
        tensors_1 = load_file(model_path_1)  # Load model 1
        tensors_2 = load_file(model_path_2)  # Load model 2

        # Find common keys between both models
        common_keys = set(tensors_1.keys()).intersection(tensors_2.keys())

        # Merging only the common tensors
        merged_tensors = {}
        for key in common_keys:
            # Ensure both tensors have the same shape before merging
            if tensors_1[key].shape == tensors_2[key].shape:
                print(f"Merging tensor: {key} (from both models)")
                merged_tensors[key] = tensors_1[key] * ratio_1 + tensors_2[key] * ratio_2
            else:
                print(f"Skipping tensor: {key} due to shape mismatch (Model 1: {tensors_1[key].shape}, Model 2: {tensors_2[key].shape})")

        # Save the merged model using save_file from safetensors.torch
        save_file(merged_tensors, output_path)  # Updated method for saving models
        print(f"Merged model saved to: {output_path}")

    except Exception as e:
        print(f"An error occurred during model merging: {e}")

# Example usage
merge_models(
    'flowgram01.safetensors',
    'diffusion_pytorch_model-00001-of-00003.safetensors',
    0.6,  # 60% for the first model
    0.4,  # 40% for the second model
    '01.safetensors'  # Output filename
)