b7625
📦 llama-cppView on GitHub →
🐛 2 fixes🔧 2 symbols
Summary
This release focuses on CUDA stability by disabling CUDA graphs when using n-cpu-moe and ensuring proper device initialization via ggml_cuda_set_device.
🐛 Bug Fixes
- CUDA: Disabled CUDA graphs when using n-cpu-moe to prevent potential issues.
- Added explicit call to ggml_cuda_set_device for better device management.
🔧 Affected Symbols
ggml_cuda_set_devicen-cpu-moe