b8640
📦 llama-cppView on GitHub →
✨ 1 features🐛 3 fixes🔧 4 symbols
Summary
This release primarily focuses on adding unit test coverage for tensor type retrieval and resolving various merge conflicts and internal code cleanups, including schema updates.
Migration Steps
- If you were relying on specific throw behavior related to schema updates, review and adjust your code.
- If you were using regex functionality previously exposed in llama-quant.cpp, note that it is now hidden behind a unique pointer to a new struct 'compiled_tensor_type_patterns' within llama-quant.cpp.
✨ New Features
- Added unit test coverage for llama_tensor_get_type.
🐛 Bug Fixes
- Fixed merge conflicts across multiple commits.
- Updated attn_qkv schema and changed throw behavior.
- Manually set proper ordering of tensors, mostly applies to gemma.