Automated account. Commits and PRs from here are made by scripts — check before merge.
Low-level, GPU and a bit of web. Box is ASUS TUF, RTX 3050 Laptop 6GB sm_86, Arch + i3 on unix.s3.
Core — C/C++ and GPU
llama.cpp/llamar.cpp— LLM inference, recurrent bits. Vulkan offload to the 3050.sharpemu— PS5 experiments.cjoka-engine— small C++ engine.dlss-moddify-3050— HMMA FP16 → IMMA INT8 + per-block quant for DLSS. No FP8 on sm_86, only FP16 and IMMA.
Web — when I want something visible fast
- React / Next / TypeScript / Tailwind. Landing pages for anime projects, dashboards, vtube overlays, art galleries.
- Rule for small stuff: it should be shippable in an evening or two and look tidy.
No merge without a build and a run on the laptop. If there is no number from the hardware, there is no claim.
- dlss-moddify-3050 — neural rendering patches for the laptop GPU.
- cjoka-engine — engine experiment.
- runya-retro — RetroKit mini canvas engine + a retro runner, procedural pixel art, zero deps, live on Pages.
- llama.cpp — fork.
Arch Linux, zsh + p10k, nvim, kitty, i3. JetBrains Mono, dark theme. One sticker on the lid per closed mod.
Quickdraw earned — more to come as PRs land.
- graaf#464 — PageRank benchmark for
Graaf_perf(closes #450). - animechan#147 — fix random quote docs example URL (fixes #145).
Steins;Gate, Spice and Wolf, K-On, Nichijou, Frieren, Bocchi, Lycoris, Kaguya. One or two episodes in the evening, not a binge.
City pop / lo-fi when reading asm. Syrniki and cocoa.
- mail: cubetitled@gmail.com
- details in RUNYA.md
