llama.cpp
CUDA/HIP: refractor mmqv to unify the calculation of nwarps and rows per block between host and device code.
#12177

Merged

Commits

refractor mmqv to unify the calculation of nwarps and rows per block between host and device code.

IMbackK committed 1 year ago
make cuda happy, as it dosent support calling host constexpr functions in device code, even though that should not be a problem.

IMbackK committed 1 year ago
Fix nits

IMbackK committed 1 year ago
Fix spelling of parameter

IMbackK committed 1 year ago
Update ggml/src/ggml-cuda/mmvq.cu

IMbackK committed 1 year ago