Aug 24 to Aug 30, 2026

CUDA-graph build for the workstation Blackwell.

A step-by-step serving recipe for running the 125B MoE on one Ryzen AI MAX+ 395 box, written by a hobbyist. Useful as a replicable baseline for that hardware.

Llama.cpp route with long context.

Upstream vLLM run with full 262K context.

Screen recording of the model reading a CLI guide and completing a task that the 27B failed.

SGLang NEXTN setup at about 37 tok/s on code.

Speculative-decoding variants benchmarked on one card.

FP8 and NVFP4 with MTP speculation on a dual-Blackwell box.

Root-cause forensics for a pipeline-parallel bug.

A benchmark comparing GGUF quantization levels on generated code, helping choose the smallest quant that still codes well.

German-language write-up of benchmark numbers, test harness and results for running the model on three second-hand 3090s.

A runtime focused on Qwen models that squeezes Flash-Next onto 16GB Macs. From the developer of the MacBat app.

Nate Herk's long course shows building client-ready bots in Grok Bot and packaging them as a paid service.

Fast demo of eleven bots including research, design and automation tasks, with the prompts used.

Official guide on how designers used bots to explore bot portraits, a notch-sized workspace and other interface ideas.