Skip to content

llama.cpp Windows Manager v3.0.0

Latest

Choose a tag to compare

@alekk89 alekk89 released this 15 Nov 14:58
3b0c879
  • Rebuilt Overview dashboard: cards flow through a responsive grid, can be dragged and resized, and Auto arrange sizes them evenly.

  • Add a Loaded models card, metric warning thresholds, stale-reading indicators, chart readouts, and Copy readings (Ctrl+C).

  • Launch settings list the values each runtime supports, including TurboQuant turbo2/turbo3/turbo4 cache types; add draft-dflash2 and draft-lilicorr.

  • Keep unsaved launch edits across refreshes; older profiles no longer ask to be saved when unchanged.

  • Load startup profiles in the background with visible progress.

  • Harden the gateway against non-local web pages, drain streams before swaps and unloads, and fix audio and image-edit routing.

  • Block custom parameters that download models, expose files or tunnels, or run tools/MCP servers.

  • Restore the previous version after a failed update; match checksums exactly and keep partial Hugging Face downloads.

  • Complete translations in every language; follow Windows theme changes; fix column resizing and editable drop-downs.

  • llwmctl reports blocked operations as failures and only guards models this Manager serves.

  • Profiles that use a newly blocked custom parameter must be edited before they load. Reinstall official CUDA WSL runtimes to add their CUDA libraries.

  • Unsigned while awaiting SignPath. From v2.7 or later, update in the app; on v2.5/v2.6, install manually and keep your data folder.