Incarnas

LLM inference in C/C++

Methodology repo for Incarnas's NVFP4 GGUF conversions with MTP head retention. Tech report + patches + bench data + reproduction scripts.

Task-dependent benchmark gap between /v1/completions and /v1/chat/completions on instruction-tuned LLMs -- two case studies on Qwen3.x GGUFs, reproduction recipe, and probes.

A Claude Code plugin that renders a multi-line status line with live usage metrics for Claude Max/Pro subscriptions.