fix: move flash imports after maybe_unpack to fix torch import errors

[deanq]

Summary

Fixes torch import errors by reordering Flash imports to occur after maybe_unpack() call.

Problem

Torch imports were failing with "No module named 'torch'" errors during execution because runpod_flash imports were happening before /app was added to sys.path.

Solution

Move runpod_flash imports after the maybe_unpack() call to ensure /app is in the path before importing modules that may have torch imports at module level.

Changes

• Reordered imports in src/lb_handler.py

Testing

• Verify torch imports work correctly
• Test load balancer handler execution
• Confirm no import errors in worker startup

Related Issues

Fixes AE-2079

[Copilot]

Pull request overview

Reorders imports in the load balancer handler to prevent Torch-related import failures by ensuring /app is added to sys.path via maybe_unpack() before importing runpod_flash.

Changes:

• Moved runpod_flash and RemoteExecutor imports to occur after maybe_unpack()
• Added inline comments explaining the required import order
• Added # noqa: E402 to suppress “import not at top of file” lint errors

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.