build: Make target improvements (#7499)
* llama: wire up builtin runner
This adds a new entrypoint into the ollama CLI to run the cgo built runner.
On Mac arm64, this will have GPU support, but on all other platforms it will
be the lowest common denominator CPU build. After we fully transition
to the new Go runners more tech-debt can be removed and we can stop building
the "default" runner via make and rely on the builtin always.
* build: Make target improvements
Add a few new targets and help for building locally.
This also adjusts the runner lookup to favor local builds, then
runners relative to the executable, and finally payloads.
* Support customized CPU flags for runners
This implements a simplified custom CPU flags pattern for the runners.
When built without overrides, the runner name contains the vector flag
we check for (AVX) to ensure we don't try to run on unsupported systems
and crash. If the user builds a customized set, we omit the naming
scheme and don't check for compatibility. This avoids checking
requirements at runtime, so that logic has been removed as well. This
can be used to build GPU runners with no vector flags, or CPU/GPU
runners with additional flags (e.g. AVX512) enabled.
* Use relative paths
If the user checks out the repo in a path that contains spaces, make gets
really confused so use relative paths for everything in-repo to avoid breakage.
* Remove payloads from main binary
* install: clean up prior libraries
This removes support for v0.3.6 and older versions (before the tar bundle)
and ensures we clean up prior libraries before extracting the bundle(s).
Without this change, runners and dependent libraries could leak when we
update and lead to subtle runtime errors.
2024-12-10 09:47:19 -08:00
|
|
|
# Generalized GPU runner build
|
|
|
|
|
|
|
|
|
|
ifndef GPU_RUNNER_NAME
|
|
|
|
|
dummy:
|
|
|
|
|
$(error This makefile is not meant to build directly, but instead included in other Makefiles that set required variables)
|
|
|
|
|
endif
|
|
|
|
|
|
2024-12-10 12:59:39 -08:00
|
|
|
GPU_GOFLAGS="-ldflags=-w -s \"-X=github.com/ollama/ollama/version.Version=$(VERSION)\" $(EXTRA_GOLDFLAGS) $(TARGET_LDFLAGS)"
|
build: Make target improvements (#7499)
* llama: wire up builtin runner
This adds a new entrypoint into the ollama CLI to run the cgo built runner.
On Mac arm64, this will have GPU support, but on all other platforms it will
be the lowest common denominator CPU build. After we fully transition
to the new Go runners more tech-debt can be removed and we can stop building
the "default" runner via make and rely on the builtin always.
* build: Make target improvements
Add a few new targets and help for building locally.
This also adjusts the runner lookup to favor local builds, then
runners relative to the executable, and finally payloads.
* Support customized CPU flags for runners
This implements a simplified custom CPU flags pattern for the runners.
When built without overrides, the runner name contains the vector flag
we check for (AVX) to ensure we don't try to run on unsupported systems
and crash. If the user builds a customized set, we omit the naming
scheme and don't check for compatibility. This avoids checking
requirements at runtime, so that logic has been removed as well. This
can be used to build GPU runners with no vector flags, or CPU/GPU
runners with additional flags (e.g. AVX512) enabled.
* Use relative paths
If the user checks out the repo in a path that contains spaces, make gets
really confused so use relative paths for everything in-repo to avoid breakage.
* Remove payloads from main binary
* install: clean up prior libraries
This removes support for v0.3.6 and older versions (before the tar bundle)
and ensures we clean up prior libraries before extracting the bundle(s).
Without this change, runners and dependent libraries could leak when we
update and lead to subtle runtime errors.
2024-12-10 09:47:19 -08:00
|
|
|
|
|
|
|
|
# TODO Unify how we handle dependencies in the dist/packaging and install flow
|
|
|
|
|
# today, cuda is bundled, but rocm is split out. Should split them each out by runner
|
|
|
|
|
DIST_GPU_RUNNER_DEPS_DIR = $(DIST_LIB_DIR)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
GPU_RUNNER_LIBS = $(wildcard $(addsuffix .$(SHARED_EXT).*,$(addprefix $(GPU_LIB_DIR)/$(SHARED_PREFIX),$(GPU_RUNNER_LIBS_SHORT))))
|
|
|
|
|
|
|
|
|
|
GPU_RUNNER_SRCS := \
|
|
|
|
|
$(filter-out $(wildcard llama/ggml-cuda/fattn*.cu),$(wildcard llama/ggml-cuda/*.cu)) \
|
|
|
|
|
$(wildcard llama/ggml-cuda/template-instances/mmq*.cu) \
|
2024-12-14 14:55:51 -08:00
|
|
|
llama/ggml.c llama/ggml-backend.cpp llama/ggml-alloc.c llama/ggml-quants.c llama/sgemm.cpp llama/ggml-threading.cpp
|
build: Make target improvements (#7499)
* llama: wire up builtin runner
This adds a new entrypoint into the ollama CLI to run the cgo built runner.
On Mac arm64, this will have GPU support, but on all other platforms it will
be the lowest common denominator CPU build. After we fully transition
to the new Go runners more tech-debt can be removed and we can stop building
the "default" runner via make and rely on the builtin always.
* build: Make target improvements
Add a few new targets and help for building locally.
This also adjusts the runner lookup to favor local builds, then
runners relative to the executable, and finally payloads.
* Support customized CPU flags for runners
This implements a simplified custom CPU flags pattern for the runners.
When built without overrides, the runner name contains the vector flag
we check for (AVX) to ensure we don't try to run on unsupported systems
and crash. If the user builds a customized set, we omit the naming
scheme and don't check for compatibility. This avoids checking
requirements at runtime, so that logic has been removed as well. This
can be used to build GPU runners with no vector flags, or CPU/GPU
runners with additional flags (e.g. AVX512) enabled.
* Use relative paths
If the user checks out the repo in a path that contains spaces, make gets
really confused so use relative paths for everything in-repo to avoid breakage.
* Remove payloads from main binary
* install: clean up prior libraries
This removes support for v0.3.6 and older versions (before the tar bundle)
and ensures we clean up prior libraries before extracting the bundle(s).
Without this change, runners and dependent libraries could leak when we
update and lead to subtle runtime errors.
2024-12-10 09:47:19 -08:00
|
|
|
GPU_RUNNER_HDRS := \
|
|
|
|
|
$(wildcard llama/ggml-cuda/*.cuh)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# Conditional flags and components to speed up developer builds
|
|
|
|
|
ifneq ($(OLLAMA_FAST_BUILD),)
|
|
|
|
|
GPU_COMPILER_CUFLAGS += \
|
|
|
|
|
-DGGML_DISABLE_FLASH_ATTN
|
|
|
|
|
else
|
|
|
|
|
GPU_RUNNER_SRCS += \
|
|
|
|
|
$(wildcard llama/ggml-cuda/fattn*.cu) \
|
|
|
|
|
$(wildcard llama/ggml-cuda/template-instances/fattn-wmma*.cu) \
|
|
|
|
|
$(wildcard llama/ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu) \
|
|
|
|
|
$(wildcard llama/ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu) \
|
|
|
|
|
$(wildcard llama/ggml-cuda/template-instances/fattn-vec*f16-f16.cu)
|
|
|
|
|
endif
|
|
|
|
|
|
|
|
|
|
GPU_RUNNER_OBJS := $(GPU_RUNNER_SRCS:.cu=.$(GPU_RUNNER_NAME).$(OBJ_EXT))
|
|
|
|
|
GPU_RUNNER_OBJS := $(GPU_RUNNER_OBJS:.c=.$(GPU_RUNNER_NAME).$(OBJ_EXT))
|
|
|
|
|
GPU_RUNNER_OBJS := $(addprefix $(BUILD_DIR)/,$(GPU_RUNNER_OBJS:.cpp=.$(GPU_RUNNER_NAME).$(OBJ_EXT)))
|
|
|
|
|
|
|
|
|
|
DIST_RUNNERS = $(addprefix $(RUNNERS_DIST_DIR)/,$(addsuffix /ollama_llama_server$(EXE_EXT),$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)))
|
|
|
|
|
BUILD_RUNNERS = $(addprefix $(RUNNERS_BUILD_DIR)/,$(addsuffix /ollama_llama_server$(EXE_EXT),$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
$(GPU_RUNNER_NAME): $(BUILD_RUNNERS)
|
|
|
|
|
|
|
|
|
|
dist: $(DIST_RUNNERS)
|
|
|
|
|
|
|
|
|
|
# Build targets
|
|
|
|
|
$(BUILD_DIR)/%.$(GPU_RUNNER_NAME).$(OBJ_EXT): %.cu
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CCACHE) $(GPU_COMPILER) -c $(GPU_COMPILER_CFLAGS) $(GPU_COMPILER_CUFLAGS) $(GPU_RUNNER_ARCH_FLAGS) -o $@ $<
|
|
|
|
|
$(BUILD_DIR)/%.$(GPU_RUNNER_NAME).$(OBJ_EXT): %.c
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CCACHE) $(GPU_COMPILER) -c $(GPU_COMPILER_CFLAGS) -o $@ $<
|
|
|
|
|
$(BUILD_DIR)/%.$(GPU_RUNNER_NAME).$(OBJ_EXT): %.cpp
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CCACHE) $(GPU_COMPILER) -c $(GPU_COMPILER_CXXFLAGS) -o $@ $<
|
|
|
|
|
$(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/ollama_llama_server$(EXE_EXT): TARGET_CGO_LDFLAGS = $(CGO_EXTRA_LDFLAGS) -L"$(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/"
|
|
|
|
|
$(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/ollama_llama_server$(EXE_EXT): $(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT) ./llama/*.go ./llama/runner/*.go $(COMMON_SRCS) $(COMMON_HDRS)
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
GOARCH=$(ARCH) CGO_LDFLAGS="$(TARGET_CGO_LDFLAGS)" go build -buildmode=pie $(GPU_GOFLAGS) -trimpath -tags $(subst $(space),$(comma),$(GPU_RUNNER_CPU_FLAGS) $(GPU_RUNNER_GO_TAGS)) -o $@ ./cmd/runner
|
|
|
|
|
$(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT): $(GPU_RUNNER_OBJS) $(COMMON_HDRS) $(GPU_RUNNER_HDRS)
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CCACHE) $(GPU_COMPILER) --shared -L$(GPU_LIB_DIR) $(GPU_RUNNER_DRIVER_LIB_LINK) -L${DIST_GPU_RUNNER_DEPS_DIR} $(foreach lib, $(GPU_RUNNER_LIBS_SHORT), -l$(lib)) $(GPU_RUNNER_OBJS) -o $@
|
|
|
|
|
|
|
|
|
|
# Distribution targets
|
|
|
|
|
$(RUNNERS_DIST_DIR)/%: $(RUNNERS_BUILD_DIR)/%
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CP) $< $@
|
|
|
|
|
$(RUNNERS_DIST_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/ollama_llama_server$(EXE_EXT): $(RUNNERS_DIST_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT) $(GPU_DIST_LIB_DEPS)
|
|
|
|
|
$(RUNNERS_DIST_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT): $(RUNNERS_BUILD_DIR)/$(GPU_RUNNER_NAME)$(GPU_RUNNER_EXTRA_VARIANT)/$(SHARED_PREFIX)ggml_$(GPU_RUNNER_NAME).$(SHARED_EXT)
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CP) $< $@
|
|
|
|
|
$(GPU_DIST_LIB_DEPS):
|
|
|
|
|
@-mkdir -p $(dir $@)
|
|
|
|
|
$(CP) $(GPU_LIB_DIR)/$(notdir $@) $(dir $@)
|
|
|
|
|
|
|
|
|
|
clean:
|
|
|
|
|
rm -f $(GPU_RUNNER_OBJS) $(BUILD_RUNNERS) $(DIST_RUNNERS)
|
|
|
|
|
|
|
|
|
|
.PHONY: clean $(GPU_RUNNER_NAME)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
# Handy debugging for make variables
|
|
|
|
|
print-%:
|
|
|
|
|
@echo '$*=$($*)'
|
|
|
|
|
|
