# Pre-compiled wheels for the HF Space (Py3.12 / linux x86_64 + cu124). # Pattern from Dean (UNRL) "Run llama.cpp on ZeroGPU" (2026-06-05): # torch ships cuda libs that llama.cpp needs at runtime → install torch # from PyTorch's cu128 index, and llama-cpp-python from the matching # cu124 prebuilt wheel index (no on-builder compilation). # 0.3.19 is the newest cp312 linux_x86_64 wheel on the index at write time. --extra-index-url https://download.pytorch.org/whl/cu128 --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu124 torch==2.8.0 transformers>=4.46.0,<5.0 accelerate sentencepiece pandas pillow spaces requests hf_transfer llama-cpp-python==0.3.19