Architecture targeting
Update build flags and project settings so the local GPU architecture is actually targeted.
Compatibility-aware repair
`cuda-doctor doctor auto`
Repair mode should feel like `npm audit fix`, but with stronger compatibility rules and a validation gate that blocks false success.
cuda-doctor doctor auto
v0.0.0Repair policy
- Inspect the current environment before changing anything.
- Prefer the newest compatible versions instead of blindly forcing latest-everything.
- Preserve the user's project and system constraints where possible.
- Make larger jumps only when the current stack cannot support the detected GPU or requested workload.
Repairs it should perform
Toolkit reconciliation
Replace or surface a CUDA toolkit that cannot support the detected GPU generation.
Framework alignment
Surface and fix mismatched PyTorch wheel and runtime pairings.
Path and module repair
Correct path and environment issues and repair common Linux driver or module mismatches.
Repair is not complete until validation passes
doctor auto should refuse to claim success based only on package presence. A repair is complete only after a real validation kernel passes on the local GPU.
Expected repair flowbash
cuda-doctor doctor auto
cuda-doctor validateRelated docs
Diagnose
doctor
Run a full environment diagnosis for the GPU, driver, toolkit, runtime, build chain, and validation risk.
Repair
setup
Prepare a machine for CUDA use from a mostly missing or incomplete state.
Execution
validate
Prove that device selection, memory transfer, kernel launch, and runtime behavior work on the local GPU.