refactor: move main close to implementation for every backend

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

Makefile

@@ -403,7 +403,7 @@ backend-assets/grpc:
 backend-assets/grpc/llama: backend-assets/grpc go-llama/libbinding.a
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-llama LIBRARY_PATH=$(shell pwd)/go-llama \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama ./cmd/grpc/llama/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama ./backend/go/llm/llama/
 # TODO: every binary should have its own folder instead, so can have different metal implementations
 ifeq ($(BUILD_TYPE),metal)
 	cp go-llama/build/bin/ggml-metal.metal backend-assets/grpc/
@@ -443,68 +443,68 @@ endif
 backend-assets/grpc/llama-stable: backend-assets/grpc go-llama-stable/libbinding.a
 	$(GOCMD) mod edit -replace github.com/go-skynet/go-llama.cpp=$(shell pwd)/go-llama-stable
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-llama-stable LIBRARY_PATH=$(shell pwd)/go-llama \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-stable ./cmd/grpc/llama-stable/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/llama-stable ./backend/go/llm/llama-stable/
 
 backend-assets/grpc/gpt4all: backend-assets/grpc backend-assets/gpt4all gpt4all/gpt4all-bindings/golang/libgpt4all.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/gpt4all/gpt4all-bindings/golang/ LIBRARY_PATH=$(shell pwd)/gpt4all/gpt4all-bindings/golang/ \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./cmd/grpc/gpt4all/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt4all ./backend/go/llm/gpt4all/
 
 backend-assets/grpc/dolly: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/dolly ./cmd/grpc/dolly/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/dolly ./backend/go/llm/dolly/
 
 backend-assets/grpc/gpt2: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt2 ./cmd/grpc/gpt2/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gpt2 ./backend/go/llm/gpt2/
 
 backend-assets/grpc/gptj: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptj ./cmd/grpc/gptj/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptj ./backend/go/llm/gptj/
 
 backend-assets/grpc/gptneox: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptneox ./cmd/grpc/gptneox/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/gptneox ./backend/go/llm/gptneox/
 
 backend-assets/grpc/mpt: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/mpt ./cmd/grpc/mpt/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/mpt ./backend/go/llm/mpt/
 
 backend-assets/grpc/replit: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/replit ./cmd/grpc/replit/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/replit ./backend/go/llm/replit/
 
 backend-assets/grpc/falcon-ggml: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/falcon-ggml ./cmd/grpc/falcon-ggml/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/falcon-ggml ./backend/go/llm/falcon-ggml/
 
 backend-assets/grpc/starcoder: backend-assets/grpc go-ggml-transformers/libtransformers.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-ggml-transformers LIBRARY_PATH=$(shell pwd)/go-ggml-transformers \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/starcoder ./cmd/grpc/starcoder/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/starcoder ./backend/go/llm/starcoder/
 
 backend-assets/grpc/rwkv: backend-assets/grpc go-rwkv/librwkv.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-rwkv LIBRARY_PATH=$(shell pwd)/go-rwkv \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./cmd/grpc/rwkv/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/rwkv ./backend/go/llm/rwkv
 
 backend-assets/grpc/bert-embeddings: backend-assets/grpc go-bert/libgobert.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-bert LIBRARY_PATH=$(shell pwd)/go-bert \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./cmd/grpc/bert-embeddings/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/bert-embeddings ./backend/go/llm/bert/
 
 backend-assets/grpc/langchain-huggingface: backend-assets/grpc
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/langchain-huggingface ./cmd/grpc/langchain-huggingface/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/langchain-huggingface ./backend/go/llm/langchain/
 
 backend-assets/grpc/stablediffusion: backend-assets/grpc
 	if [ ! -f backend-assets/grpc/stablediffusion ]; then \
 		$(MAKE) go-stable-diffusion/libstablediffusion.a; \
 		CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/go-stable-diffusion/ LIBRARY_PATH=$(shell pwd)/go-stable-diffusion/ \
-		$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./cmd/grpc/stablediffusion/; \
+		$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/stablediffusion ./backend/go/image/; \
 	fi
 
 backend-assets/grpc/piper: backend-assets/grpc backend-assets/espeak-ng-data go-piper/libpiper_binding.a
 	CGO_CXXFLAGS="$(PIPER_CGO_CXXFLAGS)" CGO_LDFLAGS="$(PIPER_CGO_LDFLAGS)" LIBRARY_PATH=$(shell pwd)/go-piper \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./cmd/grpc/piper/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/piper ./backend/go/tts/
 
 backend-assets/grpc/whisper: backend-assets/grpc whisper.cpp/libwhisper.a
 	CGO_LDFLAGS="$(CGO_LDFLAGS)" C_INCLUDE_PATH=$(shell pwd)/whisper.cpp LIBRARY_PATH=$(shell pwd)/whisper.cpp \
-	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./cmd/grpc/whisper/
+	$(GOCMD) build -ldflags "$(LD_FLAGS)" -tags "$(GO_TAGS)" -o backend-assets/grpc/whisper ./backend/go/transcribe/
 
 grpcs: prepare $(GRPC_BACKENDS)

cmd/grpc/bert-embeddings/main.go → backend/go/image/main.go

@@ -5,7 +5,6 @@ package main
 import (
 	"flag"
 
-	bert "github.com/go-skynet/LocalAI/backend/go/llm/bert"
 	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
 )
 
@@ -16,7 +15,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &bert.Embeddings{}); err != nil {
+	if err := grpc.StartServer(*addr, &StableDiffusion{}); err != nil {
 		panic(err)
 	}
 }

backend/go/image/stablediffusion.go

@@ -1,4 +1,4 @@
-package image
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

backend/go/llm/bert/bert.go

@@ -1,4 +1,4 @@
-package bert
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

cmd/grpc/rwkv/main.go → backend/go/llm/bert/main.go

@@ -5,8 +5,6 @@ package main
 import (
 	"flag"
 
-	rwkv "github.com/go-skynet/LocalAI/backend/go/llm/rwkv"
-
 	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
 )
 
@@ -17,7 +15,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &rwkv.LLM{}); err != nil {
+	if err := grpc.StartServer(*addr, &Embeddings{}); err != nil {
 		panic(err)
 	}
 }

backend/go/llm/gpt4all/gpt4all.go

@@ -1,4 +1,4 @@
-package gpt4all
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

cmd/grpc/gpt4all/main.go → backend/go/llm/gpt4all/main.go

@@ -5,8 +5,6 @@ package main
 import (
 	"flag"
 
-	gpt4all "github.com/go-skynet/LocalAI/backend/go/llm/gpt4all"
-
 	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
 )
 
@@ -17,7 +15,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &gpt4all.LLM{}); err != nil {
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
 		panic(err)
 	}
 }

backend/go/llm/langchain/langchain.go

@@ -1,4 +1,4 @@
-package langchain
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

backend/go/llm/langchain/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/llama-stable/llama.go

@@ -1,4 +1,4 @@
-package llama
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

cmd/grpc/llama-stable/main.go → backend/go/llm/llama-stable/main.go

@@ -3,8 +3,6 @@ package main
 import (
 	"flag"
 
-	llama "github.com/go-skynet/LocalAI/backend/go/llm/llama-stable"
-
 	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
 )
 
@@ -15,7 +13,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &llama.LLM{}); err != nil {
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
 		panic(err)
 	}
 }

backend/go/llm/llama/llama.go

@@ -1,4 +1,4 @@
-package llama
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

cmd/grpc/piper/main.go → backend/go/llm/llama/main.go

@@ -1,12 +1,12 @@
 package main
 
+// GRPC Falcon server
+
 // Note: this is started internally by LocalAI and a server is allocated for each model
 
 import (
 	"flag"
 
-	tts "github.com/go-skynet/LocalAI/backend/go/tts"
-
 	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
 )
 
@@ -17,7 +17,7 @@ var (
 func main() {
 	flag.Parse()
 
-	if err := grpc.StartServer(*addr, &tts.Piper{}); err != nil {
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
 		panic(err)
 	}
 }

backend/go/llm/rwkv/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &LLM{}); err != nil {
+		panic(err)
+	}
+}

backend/go/llm/rwkv/rwkv.go

@@ -1,4 +1,4 @@
-package rwkv
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

backend/go/transcribe/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &Whisper{}); err != nil {
+		panic(err)
+	}
+}

backend/go/transcribe/transcript.go

@@ -1,4 +1,4 @@
-package transcribe
+package main
 
 import (
 	"fmt"

backend/go/transcribe/whisper.go

@@ -1,4 +1,4 @@
-package transcribe
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)

backend/go/tts/main.go

@@ -0,0 +1,21 @@
+package main
+
+// Note: this is started internally by LocalAI and a server is allocated for each model
+
+import (
+	"flag"
+
+	grpc "github.com/go-skynet/LocalAI/pkg/grpc"
+)
+
+var (
+	addr = flag.String("addr", "localhost:50051", "the address to connect to")
+)
+
+func main() {
+	flag.Parse()
+
+	if err := grpc.StartServer(*addr, &Piper{}); err != nil {
+		panic(err)
+	}
+}

backend/go/tts/piper.go

@@ -1,4 +1,4 @@
-package tts
+package main
 
 // This is a wrapper to statisfy the GRPC service interface
 // It is meant to be used by the main executable that is the server for the specific backend type (falcon, gpt3, etc)