b9937
📦 llama-cppView on GitHub →
🐛 2 fixes🔧 1 symbols
Summary
This release focuses on aligning the CUDA snake fusion matcher behavior with other backends by correcting type predicates and enforcing contiguity checks for operands. Various pre-built binaries for different operating systems and hardware configurations are provided.
🐛 Bug Fixes
- CUDA: Fixed snake fusion type predicate to ensure 'a' and 'inv_b' are treated as F32, aligning with CPU and Metal backends.
- CUDA: Rejected snake fusion when operands are non-contiguous to prevent silent reading of incorrect data, mirroring guards in CPU, Vulkan, and Metal matchers.