b9890
📦 llama-cppView on GitHub →
🐛 5 fixes🔧 3 symbols
Summary
This release focuses on bug fixes across CUDA, CDNA, and BF16 logic, alongside a refactoring of the cuBLAS implementation. It also provides numerous pre-compiled binaries for different operating systems and hardware configurations.
Migration Steps
- If you rely on the CUDA build output, note that the -sm row has been removed during the cuBLAS refactoring.
🐛 Bug Fixes
- Fixed logic for CDNA + BF16 operations.
- Resolved an issue causing a bad return value.
- Corrected src0 strides and contiguous requirements.
- Fixed behavior related to GGML_CUDA_FORCE_CUBLAS.
- Fixed incorrect casts to BF16.