Metal fix

[maybites]

this pull request has two fixes:

one is arrangement of GUI

other is more substantial but untested on windows machines:

ProfilerX Apple Silicon Compatibility Fix

Problem

The ComfyUI_ProfilerX custom node was throwing errors on Apple Silicon Macs because it was using CUDA-specific PyTorch functions that don't exist on MPS (Metal Performance Shaders) backend.

Error: AttributeError: module 'torch.mps' has no attribute 'current_device'

Root Cause

The ProfilerX code contained hardcoded CUDA function calls:

• torch.cuda.reset_peak_memory_stats()
• torch.cuda.memory_allocated()
• torch.cuda.max_memory_allocated()

These functions don't exist in the MPS backend used by Apple Silicon.

Solution

Created a cross-platform memory management system:

1. Created memory_manager.py

• Detects device type (CUDA, MPS, or CPU)
• Provides unified interface for memory operations
• Gracefully handles missing functions on different platforms

2. Updated profiler_core.py

• Replaced all torch.cuda.* calls with memory manager calls
• Added fallback mechanisms for initialization failures
• Enhanced error handling and logging

3. Enhanced __init__.py

• Added try/catch blocks around ProfilerX initialization
• Gracefully disables extension if initialization fails
• Prevents ComfyUI startup crashes

4. Fixed PyTorch Installation

• Updated README to use stable PyTorch instead of CPU nightly build
• Removed confusing "Alternative" installation section

Files Modified

• custom_nodes/ComfyUI_ProfilerX/memory_manager.py (new)
• custom_nodes/ComfyUI_ProfilerX/profiler_core.py
• custom_nodes/ComfyUI_ProfilerX/__init__.py
• custom_nodes/ComfyUI_ProfilerX/tests/test_compatibility.py (new)

Testing

Run the compatibility test:

python custom_nodes/ComfyUI_ProfilerX/test_compatibility.py

Benefits

• ✅ Works on Apple Silicon (MPS)
• ✅ Works on CUDA systems
• ✅ Works on CPU-only systems
• ✅ Graceful fallback on unsupported systems
• ✅ No more startup crashes
• ✅ Better error reporting

Next Steps

1. Test the fix by running ComfyUI
2. If issues persist, disable ProfilerX: mv custom_nodes/ComfyUI_ProfilerX custom_nodes/ComfyUI_ProfilerX.disabled
3. Consider updating to official PyTorch with proper MPS support