ADFA-5187 | Dynamically size n_ctx based on model metadata and available RAM - #75

Merged

jatezzz merged 6 commits into

mainfrom

fix/ADFA-5187-dynamic-n-ctx

Aug 31, 2026

Commits on Aug 26, 2026

feat(ai-agent-local): size n_ctx per model and device
jatezzz
committed
b4efd66

Commits on Aug 27, 2026

fix(ai-agent-local): charge model weights against the KV budget
jatezzz
committed
1706b80
fix(ai-agent-local): parse the GGUF header once and fail safe on absurd sizes
jatezzz
committed
5d181b3

Commits on Aug 28, 2026

Commits on Aug 31, 2026

fix(ai-agent-local): clamp the KV product and take every model stat on IO
jatezzz
committed
ba1d14b