Observed arrival · 2026-10-30
tierlm assigns local LLM parts to memory by workload
A free, open-source engine for running local language models, placing frequently used model parts in GPU memory and the rest in system RAM.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- People running local language models on personal computers
- Worth noticing
- The page reports a 17 GB mixture-of-experts model improving from 21.7 to 33.3 tokens per second on a laptop RTX 3070.
Field notes
The page distinguishes mixture-of-experts models from dense models: it describes keeping attention, routing, and context on the GPU while selected experts can reside in RAM, whereas dense models move in whole layers. It also describes measuring settings on the user's own PC, including CPU threads and how far ahead to guess. The listed Windows release is v0.5.2; Linux currently has source-build instructions rather than a ready-made download.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue