b10645
📦 llama-cppView on GitHub →
✨ 1 features🐛 1 fixes🔧 2 symbols
Summary
Introduced the --n-cpu-ffn option for offloading FFN weights and deduplicated override loops in common code.
✨ New Features
- Add --n-cpu-ffn option to CPU-offload dense FFN weights of first N layers.
🐛 Bug Fixes
- Deduplicated --n-cpu-moe / --spec-draft-n-cpu-moe override loops in common code.