ianbarber

Set up and diagnose PyTorch ROCm environments on AMD Strix Halo (gfx1151) Linux systems. Use for selecting a supported or TheRock nightly stack, verifying real GPU kernels and attention backends, inspecting unified-memory limits, or troubleshooting invalid-device-function and out-of-memory failures.

Set up ML environments with PyTorch and auto-detect hardware. Use this when creating new ML projects, setting up PyTorch, or troubleshooting GPU/environment issues. Guides you through creating isolated conda environments with hardware-specific PyTorch builds (NVIDIA/AMD/CPU).

Log a summary of this session to Roam Research