Skip to content
← all work

Infermeld

A guarded Linux kit for running one GGUF model across AMD Vulkan and NVIDIA CUDA cards, powered by llama.cpp.

experimentalopen source
Site ↗source ↗
Infermeld on desktop
Infermeld on mobile
captured 04 Oct 2026

now

An experimental v0.1.0 source release with a guarded launcher, engine preflight and checkable evidence.

what's there

  • Mixed-vendor execution is upstream llama.cpp capability; Infermeld adds the guard rails, preflight and reproducible build instructions around it.
  • Device memory stays separate. The point is fitting a larger model, not a guaranteed speedup.
  • Every published number links to the evidence it came from.

limits

  • Experimental and Linux-only.
  • Not a fork, installer, new engine or unified memory pool.