Skip to content

v0.16.2

Choose a tag to compare

View all tags
@dadwritestech dadwritestech released this 11 Oct 11:00
· 179 commits to master since this release
v0.16.2
d69e508

Three switches on the Models tab. Everything in v0.16.0 and v0.16.1 still applies.

New

  • Real load progress. A loading model now shows loading 43% and a bar, through weights, draft and vision stages, instead of a stopwatch. The numbers come from llama.cpp's own load callback, which the router streams on /models/sse; LlamaForge listens only while something is loading. A router build without that stream keeps the old elapsed-seconds display.
  • Thinking: Auto / On / Off, plus a budget. One click in the model editor over llama.cpp's --reasoning and --reasoning-budget (unlimited, 512, 2k or 8k tokens). LlamaForge reads the GGUF's chat template to tell you whether Off can really turn thinking off (the template has a switch) or the model just thinks in <think> tags and may keep doing it. Models with no sign of reasoning don't get the row.
  • Vision on/off without deleting anything. Turning vision off frees the projector's VRAM and keeps the mmproj file on disk; turning it back on restores it. It stays off through rescans and re-downloads.

Each switch applies straight away: a loaded model is unloaded (you're asked first) and picks the change up on its next load.

Already on v0.16.x? The panel offers this update under Discover → New this week (installer copies; a git checkout updates with git pull).

Assets 2
Loading