b7677
📦 llama-cppView on GitHub →
🐛 2 fixes🔧 1 symbols
Summary
This release addresses a critical bug in Vulkan backend concerning push constant sizing for quantization, improving stability and correctness.
🐛 Bug Fixes
- Fixed an issue with incorrect push constant size for Vulkan quantize_q8_1 operations.
- Added an assert to catch further mismatches related to push constant sizes and fixed several found mismatches.
🔧 Affected Symbols
vulkan