Latest vanilla llama.cpp for gfx906

code visualizer used in the INGInious judge

OLLaMa LLM model LLaMa 3

Imagem personalizada com Ollama e modelo llama3:8b pré-instalado para uso offline

Slim ROCm 7.2.1 llama.cpp for AMD gfx906 (MI50/Vega20): multi-GPU tensor-parallel + MTP, ~1.5GB

Python3 env with integrated optimizer