ollama/llama

llama

This package integrates llama.cpp a Go package.

• CPU
• avx, avx2
• avx512
• macOS Metal
• Windows CUDA
• Windows ROCm
• Linux CUDA
• Linux ROCm

Extra build steps are required for CUDA and ROCm on Windows since nvcc and hipcc both require using msvc as the host compiler. For these small dlls are created:

• ggml-cuda.dll
• ggml-hipblas.dll

Note: it's important that memory is allocated and freed by the same compiler (e.g. msvc or mingw code). Issues from this should be rare, but there are some places where pointers are returned by the CUDA or HIP runtimes and freed elsewhere, causing a a crash. In a future change the same runtime should be used in both cases to avoid crashes.

Building

go build .

AVX

go build -tags avx .

AVX2

# go doesn't recognize `-mfma` as a valid compiler flag
# see https://github.com/golang/go/issues/17895
go env -w "CGO_CFLAGS_ALLOW=-mfma"
go env -w "CGO_CXXFLAGS_ALLOW=-mfma"
go build -tags=avx2 .

CUDA

Install the CUDA toolkit v11.3.1 then build ggml-cuda:

# builds ggml-cuda.dll
./cuda.sh

Then build this package with the cuda tag:

go build -tags=cuda .

ROCm

# builds ggml-hipblas.dll
./hipblas.sh

go build -tags=rocm .

Syncing with llama.cpp

./sync.sh ../../llama.cpp