interfaces/utils.py

import os
import gc
import shutil
import psutil
import torch


def is_disk_full(min_free_space_in_GB=10):
    total, used, free = shutil.disk_usage("/")
    free_gb = free / (1024 ** 3)
    
    if free_gb >= min_free_space_in_GB:
        print(f'enough space available ({free_gb} GB)')
        return False
    else:
        print('clean up!')
        return True
        
        
def release_model(model=None, label='Model'):
    """
    Releases CPU and GPU memory used by a model or pipeline.

    Args:
        model: The object to delete (e.g., model, pipeline).
        label: String label for log output.
    """
    using_cuda = torch.cuda.is_available()
    was_cuda = False

    # CPU memory before
    process = psutil.Process(os.getpid())
    mem_cpu_before = process.memory_info().rss / 1e6  # MB

    if using_cuda:
        mem_gpu_before = torch.cuda.memory_allocated()
        print(f"\n[{label}] GPU memory before release: {mem_gpu_before:.2f} MB")

    print(f"[{label}] CPU memory before release: {mem_cpu_before:.2f} MB")

    # Try to detect if model was on CUDA
    if model is not None:
        try:
            if hasattr(model, 'parameters'):
                was_cuda = any(p.is_cuda for p in model.parameters())
        except Exception as e:
            print(f"[{label}] Could not check device: {e}")
        del model

    # Garbage collection and cache clearing
    gc.collect()
    if using_cuda:
        if was_cuda:
            torch.cuda.empty_cache()
        else:
            print(f"[{label}] ⚠️ Model was not using CUDA, but CUDA is available.")

    # CPU memory after
    mem_cpu_after = process.memory_info().rss / 1e6  # MB
    print(f"[{label}] CPU memory after release:  {mem_cpu_after:.2f} MB")

    if using_cuda:
        mem_gpu_after = torch.cuda.memory_allocated()
        print(f"[{label}] GPU memory after release: {mem_gpu_after:.2f} MB\n")
    else:
        print(f"[{label}] CUDA not available — GPU memory not tracked.\n")