theadamdanielsson Claim this profile

Check that a GGUF quant is actually what it claims: bit-exact derivation attestations, verified across machines on public CI. Also packs quant ladders as recipes — 16 GiB -> 1.8 GiB, every file regenerated bit-exact.

Re-runnable CI evidence: default llama.cpp builds quantize differently across compilers and architectures; -ffp-contract=off on ggml-quants.c fixes it. One leg applies the llama.cpp#25353 patch to pinned master — gcc/Linux, clang/macOS and MSVC/Windows then produce the same hashes.

Obsidian plugin that quizzes you on your own notes, grades your answers, and remembers what you get wrong so sessions focus there. FSRS scheduling, your own AI key.

Valid tool calls from any local model — a drop-in OpenAI-compatible proxy for Ollama that guarantees well-formed tool calls and restores tool_choice.

Catch the LLM/AI calls you didn't need. A fast, deterministic linter that flags LLM API calls where plain code is simpler, cheaper, and more reliable. Your GPT call is a regex.

A minimalistic C++ Jinja templating engine for LLM chat templates

LLM inference in C/C++

AI first-responder for new GitHub issues: grounded, cited, humble, and silent when it isn't sure. Runs as an Action on your own API key.