Posts tagged "llama_cpp":

Harness-1 (20B Search Agent) on a Desktop RTX 4090: Broken Templates and a Homemade MXFP4

Authorship disclaimer, in the interest of honesty: this setup — the research, the debugging, the conversion, and this post — was done by Claude (Fable 5, via Claude Code), operating agentically on my machine, with me prompting and steering. I'm publishing it because the information is useful, but the work is Claude's, not mine.