Using certain naming conventions for the directory path breaks loading previous saves for LoRA Training

[cswattts]

Describe the bug

paths = []
for pattern in patterns:
    paths.extend(glob.glob(os.path.join(self.save_root, pattern)))

# Filter out non-existent paths and sort by creation time
if paths:
    paths = [p for p in paths if os.path.exists(p)]
    # remove false positives
    if '_LoRA' not in name:
        paths = [p for p in paths if '_LoRA' not in p]
    if '_refiner' not in name:
        paths = [p for p in paths if '_refiner' not in p]
    if '_t2i' not in name:
        paths = [p for p in paths if '_t2i' not in p]
    if '_cn' not in name:
        paths = [p for p in paths if '_cn' not in p]

    if len(paths) > 0:
        latest_path = max(paths, key=os.path.getctime)

This section of code in BaseSDTrainProcess.py L621:639 will remove any paths that have those suffixes in them which prevents loading of previous saves.

Recommended fix - change the check to look at the basepath and not the entire path. In the shared colab, the value in the config is set to content/output but if you modify it like I did, to for example "my_LoRA/output" then the code will not locate any previous saves.