4fadb05c50
install.sh previously set MAC_INTEL=true unconditionally when uname reported x86_64 on macOS, even after detecting Apple Silicon via sysctl. This caused ARM Mac users running under Rosetta to get the Intel llama.cpp prebuilt, skip PyTorch, and use Python 3.12. Now when Rosetta is detected, _ARCH is overridden to arm64 so the rest of the installer (Python version, torch, llama.cpp prebuilt) picks the correct architecture. Also adds matching Rosetta detection in install_llama_prebuilt.py's detect_host() so the prebuilt resolver works correctly even when invoked independently under Rosetta. |
||
|---|---|---|
| .. | ||
| backend | ||
| frontend | ||
| __init__.py | ||
| install_llama_prebuilt.py | ||
| install_python_stack.py | ||
| LICENSE.AGPL-3.0 | ||
| setup.bat | ||
| setup.ps1 | ||
| setup.sh | ||
| Unsloth_Studio_Colab.ipynb | ||