Popular repositories Loading

1. flash-next-strix-halo flash-next-strix-halo Public

   Vitronia — Qwen3.8-Flash-Next on AMD Strix Halo (gfx1151): 17 → 47 tok/s. Working MTP on ROCm, measurements, pitfalls, reproducible stack

   2

2. Vesuviuschallenge Vesuviuschallenge Public

   Vesuvius Challenge — ML pipeline for reading ancient scrolls + mesh-qa tool

   Python

3. llama.cpp llama.cpp Public

   Forked from ggml-org/llama.cpp

   LLM inference in C/C++

   C++

4. shard-scalpel shard-scalpel Public

   Vitronia — cut individual tensors out of huge sharded safetensors checkpoints over HTTP: fetch 5 GB, not 360

   Python