jul 9, 2026
2 links from the engineering internet.
github.com
aisystems
llama.cpp b9939 adds depthwise conv2d on the metal backend
build b9939 lands a metal implementation of depthwise 2d convolution (conv_2d_dw) with channel-tiled kernels and f16 support, one of several llama.cpp builds cut on the day. it speeds vision and mobile-style models that lean on depthwise convs on apple gpus.
github.com
oss
n8n 2.29.9 fixes expression isolate reuse when resolving node inputs
the 2.29.9 patch makes n8n acquire a fresh expression isolate when resolving node inputs and parameters, fixing state that could bleed between evaluations in the workflow-automation tool. a 2.30.2 pre-release ships the same fix on the newer line.