b8827
📦 llama-cppView on GitHub →
🐛 4 fixes🔧 3 symbols
Summary
This release focuses on internal refactoring within the OpenCL backend, specifically optimizing q8_0 tensor operations and dispatch logic for Adreno hardware, alongside minor whitespace fixes.
🐛 Bug Fixes
- Refactored q8_0 set_tensor and mul_mat host side dispatch for Adreno in OpenCL.
- Refactored q8_0 gemm/gemv Adreno dispatch for OpenCL.
- Refactored q8_0 set_tensor operations in OpenCL.
- Fixed whitespace issues in OpenCL code.