b9885
📦 llama-cppView on GitHub →
✨ 2 features🐛 1 fixes🔧 1 symbols
Summary
This release enables tiled matmul support on AIX CPUs, which significantly improves performance for certain models, while also addressing a segmentation fault issue on AIX by reducing stack buffer usage.
✨ New Features
- Enabled tiled matmul on ggml-cpu for AIX architecture.
- Introduced performance gains of ~2x in PP_Speed for FP32, Q4_0 and Q8_0 models when using the tiled matmul path on AIX.
🐛 Bug Fixes
- Reduced the buffer footprint for the tiled matmul path on AIX to prevent segmentation faults caused by large local stack buffers.