01The problem
A model server can fail below the model layer: kernel modules, drivers, containers and GPU access all have to agree.
02The approach
- Use official installation sources with explicit driver, toolkit and kernel-header decisions.
- Gate progress on hardware detection, nvidia-smi, container GPU access and service health.
- Provide reviewed inference presets and symptom-driven troubleshooting while keeping machine-specific credentials out of examples.
03The evidence
- The repository provides a SKILL.md, verification script and serving configuration examples.
- Its documented scope is supported Debian/Ubuntu x86_64 systems with NVIDIA GPUs.
04Scope & status
Infrastructure tooling with a defined compatibility scope. A setup recipe does not establish performance on untested hardware.
Public source is available for independent inspection.