Updated bindings

Author: djthorpe

Makefile

@@ -1,6 +1,7 @@
 # Paths to packages
 DOCKER=$(shell which docker)
 GIT=$(shell which git)
+GO=$(shell which go)
 
 # Other paths
 ROOT_PATH := $(CURDIR)
@@ -26,6 +27,16 @@ ifdef CUDA_HOME
   ONNXRUNTIME_FLAGS += --use_cuda --cuda_home=${CUDA_HOME} --cudnn_home=${CUDA_HOME}
 endif
 
+# Generate the pkg-config files
+generate: mkdir go-tidy
+	@echo "Generating pkg-config"
+	@PKG_CONFIG_PATH=${ROOT_PATH}/${BUILD_DIR} go generate ./sys/llamacpp
+
+# Test llamacpp bindings
+test: generate llamacpp
+	@echo "Running tests (sys)"
+	@PKG_CONFIG_PATH=${ROOT_PATH}/${BUILD_DIR} ${GO} test -v ./sys/llamacpp/...
+
 # Base images for building and running CUDA containers
 docker-base: docker-dep
 	@echo "Building ${DOCKER_TAG_BASE_BUILD}"
@@ -51,10 +62,10 @@ docker: docker-dep docker-base
 		--build-arg BASE_IMAGE_RUNTIME=${DOCKER_TAG_BASE_RUNTIME} \
 		-f Dockerfile.llamacpp .
 
-# Build llama-server
+# Build llama libraries
 llamacpp: submodule-checkout
 	@echo "Building llamacpp"
-	@cd llama.cpp && make -j$(nproc) libllama.a
+	@cd llama.cpp && make -j$(nproc) libllama.a libggml.a
 
 onnxruntime: submodule-checkout
 	@echo "Building onnxruntime"
@@ -95,8 +106,13 @@ mkdir:
 	@echo Mkdir ${BUILD_DIR}
 	@install -d ${BUILD_DIR}
 
+# go mod tidy
+go-tidy: go-dep
+	@echo Tidy
+	@go mod tidy
+
 # Clean
-clean: submodule-clean
+clean: submodule-clean go-tidy
 	@echo "Cleaning"
 	@rm -rf ${BUILD_DIR}
 
@@ -106,4 +122,8 @@ docker-dep:
 
 # Check for git
 git-dep:
-	@test -f "${GIT}" && test -x "${GIT}"  || (echo "Missing git binary" && exit 1)
+	@test -f "${GIT}" && test -x "${GIT}"  || (echo "Missing git binary" && exit 1)
+
+# Check for go
+go-dep:
+	@test -f "${GO}" && test -x "${GO}"  || (echo "Missing go binary" && exit 1)

build/libllamacpp.pc

@@ -0,0 +1,7 @@
+prefix=/Users/djt/projects/docker-llamacpp
+
+Name: libllamacpp.pc
+Description: No description
+Version: 0.0.0
+Cflags: -I${prefix}/llama.cpp/include
+Libs: -L${prefix}/llama.cpp -lllamacpp

build/llamacpp-darwin.pc

@@ -0,0 +1,4 @@
+Name: llamacpp-darwin.pc
+Description: No description
+Version: 0.0.0
+Libs: -framework Accelerate -framework Metal -framework Foundation -framework CoreGraphics

build/llamacpp.pc

@@ -0,0 +1,7 @@
+prefix=/Users/djt/projects/docker-llamacpp
+
+Name: llamacpp.pc
+Description: No description
+Version: 0.0.0
+Cflags: -I${prefix}/llama.cpp/include -I${prefix}/llama.cpp/ggml/include
+Libs: -L${prefix}/llama.cpp -lllama -lggml -lm -lstdc++

go.mod

@@ -0,0 +1,3 @@
+module github.com/mutablelogic/docker-llamacpp
+
+go 1.22.3

llama.cpp

@@ -0,0 +1,16 @@
+package llamacpp
+
+///////////////////////////////////////////////////////////////////////////////
+// CGO
+
+/*
+#cgo pkg-config: llamacpp
+#cgo darwin pkg-config: llamacpp-darwin
+*/
+import "C"
+
+// Generate the llamacpp pkg-config files
+// Setting the prefix to the base of the repository
+
+//go:generate go run ../pkg-config --version "0.0.0" --libs "-framework Accelerate -framework Metal -framework Foundation -framework CoreGraphics" llamacpp-darwin.pc
+//go:generate go run ../pkg-config --version "0.0.0" --prefix "../.." --cflags "-I$DOLLAR{prefix}/llama.cpp/include -I$DOLLAR{prefix}/llama.cpp/ggml/include" --libs "-L$DOLLAR{prefix}/llama.cpp -lllama -lggml -lm -lstdc++" llamacpp.pc

onnxruntime

@@ -0,0 +1,62 @@
+package llamacpp
+
+///////////////////////////////////////////////////////////////////////////////
+// CGO
+
+/*
+#cgo pkg-config: llamacpp
+#include <llama.h>
+#include <stdlib.h>
+*/
+import "C"
+import "unsafe"
+
+///////////////////////////////////////////////////////////////////////////////
+// TYPES
+
+type (
+	Ggml_numa_strategy C.enum_ggml_numa_strategy
+	Llama_model        C.struct_llama_model
+	Llama_model_params C.struct_llama_model_params
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// GLOBALS
+
+const (
+	GGML_NUMA_STRATEGY_DISABLED   = C.GGML_NUMA_STRATEGY_DISABLED
+	GGML_NUMA_STRATEGY_DISTRIBUTE = C.GGML_NUMA_STRATEGY_DISTRIBUTE
+	GGML_NUMA_STRATEGY_ISOLATE    = C.GGML_NUMA_STRATEGY_ISOLATE
+	GGML_NUMA_STRATEGY_NUMACTL    = C.GGML_NUMA_STRATEGY_NUMACTL
+	GGML_NUMA_STRATEGY_MIRROR     = C.GGML_NUMA_STRATEGY_MIRROR
+)
+
+///////////////////////////////////////////////////////////////////////////////
+// LIFECYCLE
+
+// Initialize the llama + ggml backend. If numa is true, use NUMA optimizations
+// Call once at the start of the program
+func Llama_backend_init() {
+	C.llama_backend_init()
+}
+
+func Llama_numa_init(numa Ggml_numa_strategy) {
+	C.llama_numa_init(C.enum_ggml_numa_strategy(numa))
+}
+
+// Call once at the end of the program - currently only used for MPI
+func Llama_backend_free() {
+	C.llama_backend_free()
+}
+
+// Load a model from a file
+func Llama_load_model_from_file(path string, params Llama_model_params) *Llama_model {
+	cPath := C.CString(path)
+	defer C.free(unsafe.Pointer(cPath))
+	return (*Llama_model)(C.llama_load_model_from_file(cPath, C.struct_llama_model_params(params)))
+}
+
+// Free a model
+func Llama_free_model(model *Llama_model) {
+	C.llama_free_model((*C.struct_llama_model)(model))
+}

sys/llamacpp/generate.go

@@ -0,0 +1,26 @@
+package llamacpp_test
+
+import (
+	"testing"
+
+	// Packages
+	"github.com/mutablelogic/docker-llamacpp/sys/llamacpp"
+)
+
+// 483MB
+const MODEL = "https://huggingface.co/ggerganov/ggml/resolve/main/ggml-model-gpt-2-117M.bin?download=true"
+
+func Test_llamacpp_000(t *testing.T) {
+	llamacpp.Llama_backend_free()
+	llamacpp.Llama_backend_free()
+}
+
+func Test_llamacpp_001(t *testing.T) {
+	llamacpp.Llama_backend_free()
+	t.Cleanup(func() {
+		llamacpp.Llama_backend_free()
+	})
+
+	model := llamacpp.Llama_load_model_from_file("/private/tmp/ggml-model-gpt-2-117M.bin", llamacpp.Llama_model_params{})
+	llamacpp.Llama_free_model(model)
+}