Pinned Loading

1. slidewright slidewright Public

   Editor visuale locale per deck di slide HTML — riordino drag&drop, editing testo/grafico (mini-PowerPoint), export HTML + PDF 16:9. 100% locale, zero dipendenze a runtime.

   JavaScript 8

2. capy-flash capy-flash Public

   Forked from Neroued/ninfer

   Qwen3.8-Flash-Next (MoE) and Qwen3.8-27B on one or two 16 GB RTX 50-series GPUs: experts offloaded to CPU, two-GPU tensor parallel. 64 GB of RAM is enough for Flash-Next. Formerly ninfer-tp2.

   C++ 93 14

3. vramble vramble Public

   One GPU, many local services that each want it whole. vramble hands out one lease at a time, queues the rest with ids and priorities, and tells whoever arrives second who is holding the card and fo…

   Python 9 1

4. wattwright wattwright Public

   Measure your own GPU clock/watt/token curve for LLM inference, then run at the point you actually want

   Python 8

5. bitbop bitbop Public

   BitBop: latent-free ternary training of small language models

   Python 6 1