jul 18, 2026
3 links from the engineering internet.
github.com
ai
llama.cpp b10068 fixes injected k/v cache rotation for dflash
the b10068 build rotates the injected k/v cache when k/v quantization is on in the dflash model path, fixing wrong context reuse for quantized kv caches.
github.com
ai
openai codex 0.144.6 corrects gpt-5.6 context window to 272k tokens
the 0.144.6 patch refreshes bundled instructions for the gpt-5.6 models and sets their context window to 272,000 tokens, so codex stops cutting long sessions short.
github.com
oss
pnpm 11.15 resolves optional peer deps from the existing graph
pnpm 11.15.0 now resolves optional peer deps declared only via peerdependenciesmeta from a version already in the graph, bumps adm-zip against a zip memory bomb, and fixes pnpm version -r on empty intents.