chore(cuda): reduce binary size (#3379)

fix(cuda): reduce binary size

Signed-off-by: Ettore Di Giacinto <[email protected]>

Author: mudler

Dockerfile

@@ -286,7 +286,14 @@ COPY --from=grpc /opt/grpc /usr/local
 WORKDIR /build
 
 ## Build the binary
-RUN make build
+## If it's CUDA, we want to skip some of the llama-compat backends to save space
+## We only leave the most CPU-optimized variant and the fallback for the cublas build
+## (both will use CUDA for the actual computation)
+RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \
+        SKIP_GRPC_BACKEND="backend-assets/grpc/llama-cpp-avx backend-assets/grpc/llama-cpp-avx2" make build; \
+    else \
+        make build; \
+    fi
 
 RUN if [ ! -d "/build/sources/go-piper/piper-phonemize/pi/lib/" ]; then \
         mkdir -p /build/sources/go-piper/piper-phonemize/pi/lib/ \

Makefile

@@ -8,7 +8,7 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=3ba780e2a8f0ffe13f571b27f0bbf2ca5a199efc
+CPPLLAMA_VERSION?=e11bd856d538e44d24d8cad4b0381fba0984d162
 
 # go-rwkv version
 RWKV_REPO?=https://github.com/donomii/go-rwkv.cpp