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

CUDA: quantized KV support for FA vec (#7527)

Browse source 
* CUDA: quantized KV support for FA vec

* try CI fix

* fix commented-out kernel variants

* add q8_0 q4_0 tests

* fix nwarps > batch size

* split fattn compile via extern templates

* fix flake8

* fix metal tests

* fix cmake

* make generate_cu_files.py executable

* add autogenerated .cu files

* fix AMD

* error if type_v != FP16 and not flash_attn

* remove obsolete code
This commit is contained in:
Johannes Gäßler 2024-06-01 08:44:14 +02:00 • committed by GitHub
parent a323ec60af
commit 9b596417af
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
110 changed files with 2697 additions and 1200 deletions
Download patch file Download diff file Expand all files Collapse all files

21
Makefile
View file

@ -421,6 +421,15 @@ ifdef LLAMA_CUBLAS
LLAMA_CUDA := 1
endif
OBJS_CUDA_TEMP_INST = $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-wmma*.cu))
ifdef LLAMA_CUDA_FA_ALL_QUANTS
OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*.cu))
else
OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu))
OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu))
OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*f16-f16.cu))
endif # LLAMA_CUDA_FA_ALL_QUANTS
ifdef LLAMA_CUDA
ifneq ('', '$(wildcard /opt/cuda)')
CUDA_PATH ?= /opt/cuda
@ -431,6 +440,7 @@ ifdef LLAMA_CUDA
MK_LDFLAGS += -lcuda -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L$(CUDA_PATH)/lib64 -L/usr/lib64 -L$(CUDA_PATH)/targets/$(UNAME_M)-linux/lib -L/usr/lib/wsl/lib
OBJS += ggml-cuda.o
OBJS += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/*.cu))
OBJS += $(OBJS_CUDA_TEMP_INST)
MK_NVCCFLAGS += -use_fast_math
ifdef LLAMA_FATAL_WARNINGS
MK_NVCCFLAGS += -Werror all-warnings
@ -493,7 +503,10 @@ ifdef LLAMA_CUDA_NO_PEER_COPY
endif # LLAMA_CUDA_NO_PEER_COPY
ifdef LLAMA_CUDA_CCBIN
MK_NVCCFLAGS += -ccbin $(LLAMA_CUDA_CCBIN)
endif
endif # LLAMA_CUDA_CCBIN
ifdef LLAMA_CUDA_FA_ALL_QUANTS
MK_NVCCFLAGS += -DGGML_CUDA_FA_ALL_QUANTS
endif # LLAMA_CUDA_FA_ALL_QUANTS
ifdef JETSON_EOL_MODULE_DETECT
define NVCC_COMPILE
@ -505,7 +518,7 @@ define NVCC_COMPILE
endef # NVCC_COMPILE
endif # JETSON_EOL_MODULE_DETECT
ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/common.cuh
ggml-cuda/%.o: ggml-cuda/%.cu ggml.h ggml-common.h ggml-cuda/common.cuh
$(NVCC_COMPILE)
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
@ -585,11 +598,12 @@ ifdef LLAMA_CUDA_NO_PEER_COPY
endif # LLAMA_CUDA_NO_PEER_COPY
OBJS += ggml-cuda.o
OBJS += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/*.cu))
OBJS += $(OBJS_CUDA_TEMP_INST)
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml.h ggml-backend.h ggml-backend-impl.h ggml-common.h $(wildcard ggml-cuda/*.cuh)
$(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
ggml-cuda/%.o: ggml-cuda/%.cu ggml-cuda/%.cuh ggml.h ggml-common.h ggml-cuda/common.cuh
ggml-cuda/%.o: ggml-cuda/%.cu ggml.h ggml-common.h ggml-cuda/common.cuh
$(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
endif # LLAMA_HIPBLAS
@ -749,6 +763,7 @@ libllama.a: llama.o ggml.o $(OBJS) $(COMMON_DEPS)
clean:
rm -vrf *.o tests/*.o *.so *.a *.dll benchmark-matmult lookup-create lookup-merge lookup-stats common/build-info.cpp *.dot $(COV_TARGETS) $(BUILD_TARGETS) $(TEST_TARGETS)
rm -vrf ggml-cuda/*.o
rm -vrf ggml-cuda/template-instances/*.o
find examples pocs -type f -name "*.o" -delete
#