Change8

b9010

📦 llama-cppView on GitHub →
1 features🐛 1 fixes🔧 1 symbols

Summary

This release fixes a critical bug in CUDA device de-duplication that could lead to OOM errors and GPU misidentification. It also introduces support for HIP and MUSA macros.

✨ New Features

  • Added HIP and MUSA macros support.

🐛 Bug Fixes

  • Fixed CUDA device PCI bus ID detection logic to prevent Out-Of-Memory (OOM) errors when de-duplicating devices, which previously caused ignoring other GPUs entirely.

Affected Symbols