[Bug] Default device hardcoded to 'cuda' causes RuntimeError on CPU-only machines.

[NANDA2726]

The default device in app.py is hardcoded to "cuda":

st.session_state.setdefault("device", "cuda")

This causes a RuntimeError when users try to load a model on a CPU-only machine, even though the GUI explicitly supports cpu, cuda, and mps as device options in the sidebar dropdown.
The app is expected to dynamically detect the available device at runtime instead of assuming CUDA is always available.

Proposed Fix

if "device" not in st.session_state:
    if torch.cuda.is_available():
        default_device = "cuda"
    elif torch.backends.mps.is_available():
        default_device = "mps"
    else:
        default_device = "cpu"
        st.session_state.device = default_device

[princedhingra183254-png]

Initialize session state if it's the first run

if "device" not in st.session_state:
st.session_state["device"] = detect_hardware()

Sidebar selection for manual override

available_options = ["cpu"]
if torch.cuda.is_available():
available_options.append("cuda")
if hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
available_options.append("mps")

Use the detected device as the default index in your sidebar

st.session_state["device"] = st.sidebar.selectbox(
"Inference Device",
options=available_options,
index=available_options.index(st.session_state["device"])
)