b10197
📦 llama-cppView on GitHub →
✨ 5 features🐛 3 fixes🔧 3 symbols
Summary
This release introduces enhanced support for alternative convolution layouts, including graph building time checks and specific kernel extensions. It also addresses several bug fixes related to test cases and Vulkan backend operations.
✨ New Features
- Added support for alternative convolution layouts in test cases.
- Introduced graph building time layout checks for convolutions.
- Extended layout checks for conv2d.cu kernel.
- Added a new graph build-time check in ggml_backend_vk_device_supports_op to fix CI failures and Vulkan run-time assert failures.
- Added an additional check in the support_op function for Vulkan to fix run-time assert failures.
🐛 Bug Fixes
- Ensured CPU back-end kernels are stored contiguously to prevent test failures with cwhn=1.
- Fixed CI failure and Vulkan run-time assert failure by introducing a new graph build-time check in ggml_backend_vk_device_supports_op.
- Fixed run-time assert failure in Vulkan backend by adding an additional check in the support_op function.