VRAM Memory not released correctly generating different images in sequence - SOLVED

[RetroUpgrade-YouTube]

I am encountering that the VRAM usage increases every time i pick a new image and run the generation step it does not release the last cached output and keeps building up until it goes out of memory and starts using CPU&RAM instad .

current setup RTX-3090, 64gb RAM, Ryzen 5900X

example of the issue >
First run snowman : total 14.2GB VRAM used
Second run Dragon : Total 17.2GB VRAM used
Third Run Snowman Again : Total Vram 23.9gb VRAM used
forth Random image: model switches to CPU usage on ram instead 27.5GB used and stays used .

already fixed this localy , not really used to git hub so dont know how to send a change for review so posting the solution i made here.

In app.py:
In the def generate_3d()
exactly before the return function add this should end like this :

del outputs  # Explicitly delete output tensors
torch.cuda.empty_cache()  # Force release unused memory

return normal_image_display, mesh_path, mesh_path

the lines i added are

    del outputs  # Explicitly delete output tensors
    torch.cuda.empty_cache()  # Force release unused memory

here is the whole function for ppl to copy past in :

def generate_3d(image, seed=-1,  
                ss_guidance_strength=3, ss_sampling_steps=50,
                slat_guidance_strength=3, slat_sampling_steps=6,):
    if image is None:
        return None, None, None

    if seed == -1:
        seed = np.random.randint(0, MAX_SEED)
    
    image = hi3dgen_pipeline.preprocess_image(image, resolution=1024)
    normal_image = normal_predictor(image, resolution=768, match_input_resolution=True, data_type='object')

    outputs = hi3dgen_pipeline.run(
        normal_image,
        seed=seed,
        formats=["mesh",],
        preprocess_image=False,
        sparse_structure_sampler_params={
            "steps": ss_sampling_steps,
            "cfg_strength": ss_guidance_strength,
        },
        slat_sampler_params={
            "steps": slat_sampling_steps,
            "cfg_strength": slat_guidance_strength,
        },
    )
    generated_mesh = outputs['mesh'][0]
    
    # Save outputs
    import datetime
    output_id = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
    os.makedirs(os.path.join(TMP_DIR, output_id), exist_ok=True)
    mesh_path = f"{TMP_DIR}/{output_id}/mesh.glb"
    
    # Export mesh
    trimesh_mesh = generated_mesh.to_trimesh(transform_pose=True)

    trimesh_mesh.export(mesh_path)
    del outputs  # Explicitly delete output tensors
    torch.cuda.empty_cache()  # Force release unused memory
    return normal_image, mesh_path, mesh_path