vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Improve cuBLAS performance by dequantizing on the GPU (#1065)

Browse source
This commit is contained in:
slaren 2023-04-20 03:14:14 +02:00 • committed by GitHub
parent 834695fe3a
commit 02d6988121
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
5 changed files with 221 additions and 41 deletions
Download patch file Download diff file Expand all files Collapse all files

24
Makefile
View file

@ -1,3 +1,6 @@
# Define the default target now so that it is always the first target
default: main quantize quantize-stats perplexity embedding vdot
ifndef UNAME_S
UNAME_S := $(shell uname -s)
endif
@ -100,6 +103,9 @@ endif
ifdef LLAMA_CUBLAS
CFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include
LDFLAGS += -lcublas_static -lculibos -lcudart_static -lcublasLt_static -lpthread -ldl -L/usr/local/cuda/lib64
OBJS += ggml-cuda.o
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
nvcc -arch=native -c -o $@ $<
endif
ifdef LLAMA_GPROF
CFLAGS += -pg
@ -137,8 +143,6 @@ $(info I CC: $(CCV))
$(info I CXX: $(CXXV))
$(info )
default: main quantize quantize-stats perplexity embedding vdot
#
# Build library
#
@ -155,35 +159,35 @@ common.o: examples/common.cpp examples/common.h
clean:
rm -vf *.o main quantize quantize-stats perplexity embedding benchmark-q4_0-matmult
main: examples/main/main.cpp ggml.o llama.o common.o
main: examples/main/main.cpp ggml.o llama.o common.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
@echo
@echo '==== Run ./main -h for help. ===='
@echo
quantize: examples/quantize/quantize.cpp ggml.o llama.o
quantize: examples/quantize/quantize.cpp ggml.o llama.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
quantize-stats: examples/quantize-stats/quantize-stats.cpp ggml.o llama.o
quantize-stats: examples/quantize-stats/quantize-stats.cpp ggml.o llama.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
perplexity: examples/perplexity/perplexity.cpp ggml.o llama.o common.o
perplexity: examples/perplexity/perplexity.cpp ggml.o llama.o common.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
embedding: examples/embedding/embedding.cpp ggml.o llama.o common.o
embedding: examples/embedding/embedding.cpp ggml.o llama.o common.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
vdot: pocs/vdot/vdot.cpp ggml.o
vdot: pocs/vdot/vdot.cpp ggml.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
libllama.so: llama.o ggml.o
libllama.so: llama.o ggml.o $(OBJS)
$(CXX) $(CXXFLAGS) -shared -fPIC -o $@ $^ $(LDFLAGS)
#
# Tests
#
benchmark: examples/benchmark/benchmark-q4_0-matmult.c ggml.o
benchmark: examples/benchmark/benchmark-q4_0-matmult.c ggml.o $(OBJS)
$(CXX) $(CXXFLAGS) $^ -o benchmark-q4_0-matmult $(LDFLAGS)
./benchmark-q4_0-matmult