b9848
📦 llama-cppView on GitHub →
🐛 1 fixes🔧 1 symbols
Summary
This release primarily addresses a bug in CUDA handling for large tables and provides numerous pre-compiled binaries across different operating systems and hardware accelerators.
🐛 Bug Fixes
- Fixed get_rows_back functionality for CUDA tables exceeding 65535 rows by implementing grid-y clamp and stride adjustments.