feat(llama.cpp): Enable decentralized, distributed inference

As #2324 introduced distributed inferencing thanks to
@rgerganov implementation in ggerganov/llama.cpp#6829 in upstream llama.cpp, now
it is possible to distribute the workload to remote llama.cpp gRPC server.

This changeset now uses mudler/edgevpn to establish a secure, distributed network between the nodes using a shared token.
The token is generated automatically when starting the server with the `--p2p` flag, and can be used by starting the workers
with `local-ai worker p2p-llama-cpp-rpc` by passing the token via environment variable (TOKEN) or with args (--token).

As per how mudler/edgevpn works, a network is established between the server and the workers with dht and mdns discovery protocols,
the llama.cpp rpc server is automatically started and exposed to the underlying p2p network so the API server can connect on.

When the HTTP server is started, it will discover the workers in the network and automatically create the port-forwards to the service locally.
Then llama.cpp is configured to use the services.

This feature is behind the "p2p" GO_FLAGS

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

core/cli/cli.go

@@ -1,21 +1,16 @@
 package cli
 
-import "embed"
-
-type Context struct {
-	Debug    bool    `env:"LOCALAI_DEBUG,DEBUG" default:"false" hidden:"" help:"DEPRECATED, use --log-level=debug instead. Enable debug logging"`
-	LogLevel *string `env:"LOCALAI_LOG_LEVEL" enum:"error,warn,info,debug,trace" help:"Set the level of logs to output [${enum}]"`
-
-	// This field is not a command line argument/flag, the struct tag excludes it from the parsed CLI
-	BackendAssets embed.FS `kong:"-"`
-}
+import (
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
+	"github.com/go-skynet/LocalAI/core/cli/worker"
+)
 
 var CLI struct {
-	Context `embed:""`
+	cliContext.Context `embed:""`
 
-	Run            RunCMD            `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
-	Models         ModelsCMD         `cmd:"" help:"Manage LocalAI models and definitions"`
-	TTS            TTSCMD            `cmd:"" help:"Convert text to speech"`
-	Transcript     TranscriptCMD     `cmd:"" help:"Convert audio to text"`
-	LLAMACPPWorker LLAMACPPWorkerCMD `cmd:"" help:"Run workers to distribute workload (llama.cpp-only)"`
+	Run        RunCMD        `cmd:"" help:"Run LocalAI, this the default command if no other command is specified. Run 'local-ai run --help' for more information" default:"withargs"`
+	Models     ModelsCMD     `cmd:"" help:"Manage LocalAI models and definitions"`
+	TTS        TTSCMD        `cmd:"" help:"Convert text to speech"`
+	Transcript TranscriptCMD `cmd:"" help:"Convert audio to text"`
+	Worker     worker.Worker `cmd:"" help:"Run workers to distribute workload (llama.cpp-only)"`
 }

core/cli/context/context.go

@@ -0,0 +1,11 @@
+package cliContext
+
+import "embed"
+
+type Context struct {
+	Debug    bool    `env:"LOCALAI_DEBUG,DEBUG" default:"false" hidden:"" help:"DEPRECATED, use --log-level=debug instead. Enable debug logging"`
+	LogLevel *string `env:"LOCALAI_LOG_LEVEL" enum:"error,warn,info,debug,trace" help:"Set the level of logs to output [${enum}]"`
+
+	// This field is not a command line argument/flag, the struct tag excludes it from the parsed CLI
+	BackendAssets embed.FS `kong:"-"`
+}

core/cli/models.go

@@ -4,6 +4,8 @@ import (
 	"encoding/json"
 	"fmt"
 
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
+
 	"github.com/go-skynet/LocalAI/pkg/gallery"
 	"github.com/rs/zerolog/log"
 	"github.com/schollz/progressbar/v3"
@@ -29,7 +31,7 @@ type ModelsCMD struct {
 	Install ModelsInstall `cmd:"" help:"Install a model from the gallery"`
 }
 
-func (ml *ModelsList) Run(ctx *Context) error {
+func (ml *ModelsList) Run(ctx *cliContext.Context) error {
 	var galleries []gallery.Gallery
 	if err := json.Unmarshal([]byte(ml.Galleries), &galleries); err != nil {
 		log.Error().Err(err).Msg("unable to load galleries")
@@ -49,7 +51,7 @@ func (ml *ModelsList) Run(ctx *Context) error {
 	return nil
 }
 
-func (mi *ModelsInstall) Run(ctx *Context) error {
+func (mi *ModelsInstall) Run(ctx *cliContext.Context) error {
 	modelName := mi.ModelArgs[0]
 
 	var galleries []gallery.Gallery

core/cli/run.go

@@ -1,12 +1,15 @@
 package cli
 
 import (
+	"context"
 	"fmt"
 	"strings"
 	"time"
 
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
 	"github.com/go-skynet/LocalAI/core/config"
 	"github.com/go-skynet/LocalAI/core/http"
+	"github.com/go-skynet/LocalAI/core/p2p"
 	"github.com/go-skynet/LocalAI/core/startup"
 	"github.com/rs/zerolog"
 	"github.com/rs/zerolog/log"
@@ -37,13 +40,14 @@ type RunCMD struct {
 	Threads     int  `env:"LOCALAI_THREADS,THREADS" short:"t" default:"4" help:"Number of threads used for parallel computation. Usage of the number of physical cores in the system is suggested" group:"performance"`
 	ContextSize int  `env:"LOCALAI_CONTEXT_SIZE,CONTEXT_SIZE" default:"512" help:"Default context size for models" group:"performance"`
 
-	Address          string   `env:"LOCALAI_ADDRESS,ADDRESS" default:":8080" help:"Bind address for the API server" group:"api"`
-	CORS             bool     `env:"LOCALAI_CORS,CORS" help:"" group:"api"`
-	CORSAllowOrigins string   `env:"LOCALAI_CORS_ALLOW_ORIGINS,CORS_ALLOW_ORIGINS" group:"api"`
-	UploadLimit      int      `env:"LOCALAI_UPLOAD_LIMIT,UPLOAD_LIMIT" default:"15" help:"Default upload-limit in MB" group:"api"`
-	APIKeys          []string `env:"LOCALAI_API_KEY,API_KEY" help:"List of API Keys to enable API authentication. When this is set, all the requests must be authenticated with one of these API keys" group:"api"`
-	DisableWebUI     bool     `env:"LOCALAI_DISABLE_WEBUI,DISABLE_WEBUI" default:"false" help:"Disable webui" group:"api"`
-
+	Address              string   `env:"LOCALAI_ADDRESS,ADDRESS" default:":8080" help:"Bind address for the API server" group:"api"`
+	CORS                 bool     `env:"LOCALAI_CORS,CORS" help:"" group:"api"`
+	CORSAllowOrigins     string   `env:"LOCALAI_CORS_ALLOW_ORIGINS,CORS_ALLOW_ORIGINS" group:"api"`
+	UploadLimit          int      `env:"LOCALAI_UPLOAD_LIMIT,UPLOAD_LIMIT" default:"15" help:"Default upload-limit in MB" group:"api"`
+	APIKeys              []string `env:"LOCALAI_API_KEY,API_KEY" help:"List of API Keys to enable API authentication. When this is set, all the requests must be authenticated with one of these API keys" group:"api"`
+	DisableWebUI         bool     `env:"LOCALAI_DISABLE_WEBUI,DISABLE_WEBUI" default:"false" help:"Disable webui" group:"api"`
+	Peer2Peer            bool     `env:"LOCALAI_P2P,P2P" name:"p2p" default:"false" help:"Enable P2P mode" group:"p2p"`
+	Peer2PeerToken       string   `env:"LOCALAI_P2P_TOKEN,P2P_TOKEN" name:"p2ptoken" help:"Token for P2P mode (optional)" group:"p2p"`
 	ParallelRequests     bool     `env:"LOCALAI_PARALLEL_REQUESTS,PARALLEL_REQUESTS" help:"Enable backends to handle multiple requests in parallel if they support it (e.g.: llama.cpp or vllm)" group:"backends"`
 	SingleActiveBackend  bool     `env:"LOCALAI_SINGLE_ACTIVE_BACKEND,SINGLE_ACTIVE_BACKEND" help:"Allow only one backend to be run at a time" group:"backends"`
 	PreloadBackendOnly   bool     `env:"LOCALAI_PRELOAD_BACKEND_ONLY,PRELOAD_BACKEND_ONLY" default:"false" help:"Do not launch the API services, only the preloaded models / backends are started (useful for multi-node setups)" group:"backends"`
@@ -54,7 +58,7 @@ type RunCMD struct {
 	WatchdogBusyTimeout  string   `env:"LOCALAI_WATCHDOG_BUSY_TIMEOUT,WATCHDOG_BUSY_TIMEOUT" default:"5m" help:"Threshold beyond which a busy backend should be stopped" group:"backends"`
 }
 
-func (r *RunCMD) Run(ctx *Context) error {
+func (r *RunCMD) Run(ctx *cliContext.Context) error {
 	opts := []config.AppOption{
 		config.WithConfigFile(r.ModelsConfigFile),
 		config.WithJSONStringPreload(r.PreloadModels),
@@ -81,6 +85,27 @@ func (r *RunCMD) Run(ctx *Context) error {
 		config.WithModelsURL(append(r.Models, r.ModelArgs...)...),
 	}
 
+	if r.Peer2Peer || r.Peer2PeerToken != "" {
+		log.Info().Msg("P2P mode enabled")
+		token := r.Peer2PeerToken
+		if token == "" {
+			// IF no token is provided, and p2p is enabled,
+			// we generate one and wait for the user to pick up the token (this is for interactive)
+			log.Info().Msg("No token provided, generating one")
+			token = p2p.GenerateToken()
+			log.Info().Msg("Generated Token:")
+			fmt.Println(token)
+			// Ask for user confirmation
+			log.Info().Msg("Press a button to proceed")
+			var input string
+			fmt.Scanln(&input)
+		}
+		log.Info().Msg("Starting P2P server discovery...")
+		if err := p2p.LLamaCPPRPCServerDiscoverer(context.Background(), token); err != nil {
+			return err
+		}
+	}
+
 	idleWatchDog := r.EnableWatchdogIdle
 	busyWatchDog := r.EnableWatchdogBusy
 

core/cli/transcript.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 
 	"github.com/go-skynet/LocalAI/core/backend"
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
 	"github.com/go-skynet/LocalAI/core/config"
 	"github.com/go-skynet/LocalAI/pkg/model"
 	"github.com/rs/zerolog/log"
@@ -22,7 +23,7 @@ type TranscriptCMD struct {
 	BackendAssetsPath string `env:"LOCALAI_BACKEND_ASSETS_PATH,BACKEND_ASSETS_PATH" type:"path" default:"/tmp/localai/backend_data" help:"Path used to extract libraries that are required by some of the backends in runtime" group:"storage"`
 }
 
-func (t *TranscriptCMD) Run(ctx *Context) error {
+func (t *TranscriptCMD) Run(ctx *cliContext.Context) error {
 	opts := &config.ApplicationConfig{
 		ModelPath:         t.ModelsPath,
 		Context:           context.Background(),

core/cli/tts.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 
 	"github.com/go-skynet/LocalAI/core/backend"
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
 	"github.com/go-skynet/LocalAI/core/config"
 	"github.com/go-skynet/LocalAI/pkg/model"
 	"github.com/rs/zerolog/log"
@@ -24,7 +25,7 @@ type TTSCMD struct {
 	BackendAssetsPath string `env:"LOCALAI_BACKEND_ASSETS_PATH,BACKEND_ASSETS_PATH" type:"path" default:"/tmp/localai/backend_data" help:"Path used to extract libraries that are required by some of the backends in runtime" group:"storage"`
 }
 
-func (t *TTSCMD) Run(ctx *Context) error {
+func (t *TTSCMD) Run(ctx *cliContext.Context) error {
 	outputFile := t.OutputFile
 	outputDir := t.BackendAssetsPath
 	if outputFile != "" {

core/cli/worker/worker.go

@@ -0,0 +1,10 @@
+package worker
+
+type WorkerFlags struct {
+	BackendAssetsPath string `env:"LOCALAI_BACKEND_ASSETS_PATH,BACKEND_ASSETS_PATH" type:"path" default:"/tmp/localai/backend_data" help:"Path used to extract libraries that are required by some of the backends in runtime" group:"storage"`
+}
+
+type Worker struct {
+	P2P      P2P      `cmd:"" name:"p2p-llama-cpp-rpc" help:"Starts a LocalAI llama.cpp worker in P2P mode (requires a token)"`
+	LLamaCPP LLamaCPP `cmd:"" name:"llama-cpp-rpc" help:"Starts a llama.cpp worker in standalone mode"`
+}

core/cli/llamacppworker.go → core/cli/worker/worker_llamacpp.go

@@ -1,26 +1,32 @@
-package cli
+package worker
 
 import (
+	"fmt"
 	"os"
 	"syscall"
 
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
 	"github.com/go-skynet/LocalAI/pkg/assets"
 	"github.com/rs/zerolog/log"
 )
 
-type LLAMACPPWorkerCMD struct {
-	Args              []string `arg:"" optional:"" name:"models" help:"Worker arguments: host port"`
-	BackendAssetsPath string   `env:"LOCALAI_BACKEND_ASSETS_PATH,BACKEND_ASSETS_PATH" type:"path" default:"/tmp/localai/backend_data" help:"Path used to extract libraries that are required by some of the backends in runtime" group:"storage"`
+type LLamaCPP struct {
+	Args        []string `arg:"" optional:"" name:"models" help:"Model configuration URLs to load"`
+	WorkerFlags `embed:""`
 }
 
-func (r *LLAMACPPWorkerCMD) Run(ctx *Context) error {
+func (r *LLamaCPP) Run(ctx *cliContext.Context) error {
 	// Extract files from the embedded FS
 	err := assets.ExtractFiles(ctx.BackendAssets, r.BackendAssetsPath)
 	log.Debug().Msgf("Extracting backend assets files to %s", r.BackendAssetsPath)
 	if err != nil {
 		log.Warn().Msgf("Failed extracting backend assets files: %s (might be required for some backends to work properly, like gpt4all)", err)
 	}
 
+	if len(os.Args) < 4 {
+		return fmt.Errorf("usage: local-ai worker llama-cpp-rpc -- <llama-rpc-server-args>")
+	}
+
 	return syscall.Exec(
 		assets.ResolvePath(
 			r.BackendAssetsPath,
@@ -32,6 +38,6 @@ func (r *LLAMACPPWorkerCMD) Run(ctx *Context) error {
 				r.BackendAssetsPath,
 				"util",
 				"llama-cpp-rpc-server",
-			)}, r.Args...),
+			)}, os.Args[4:]...),
 		os.Environ())
 }

core/cli/worker/worker_nop2p.go

@@ -0,0 +1,16 @@
+//go:build !p2p
+// +build !p2p
+
+package worker
+
+import (
+	"fmt"
+
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
+)
+
+type P2P struct{}
+
+func (r *P2P) Run(ctx *cliContext.Context) error {
+	return fmt.Errorf("p2p mode is not enabled in this build")
+}

core/cli/worker/worker_p2p.go

@@ -0,0 +1,104 @@
+//go:build p2p
+// +build p2p
+
+package worker
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"os/exec"
+	"time"
+
+	cliContext "github.com/go-skynet/LocalAI/core/cli/context"
+	"github.com/go-skynet/LocalAI/core/p2p"
+	"github.com/go-skynet/LocalAI/pkg/assets"
+	"github.com/phayes/freeport"
+	"github.com/rs/zerolog/log"
+)
+
+type P2P struct {
+	WorkerFlags       `embed:""`
+	Token             string   `env:"LOCALAI_TOKEN,TOKEN" help:"JSON list of galleries"`
+	NoRunner          bool     `env:"LOCALAI_NO_RUNNER,NO_RUNNER" help:"Do not start the llama-cpp-rpc-server"`
+	RunnerAddress     string   `env:"LOCALAI_RUNNER_ADDRESS,RUNNER_ADDRESS" help:"Address of the llama-cpp-rpc-server"`
+	RunnerPort        string   `env:"LOCALAI_RUNNER_PORT,RUNNER_PORT" help:"Port of the llama-cpp-rpc-server"`
+	ExtraLLamaCPPArgs []string `env:"LOCALAI_EXTRA_LLAMA_CPP_ARGS,EXTRA_LLAMA_CPP_ARGS" help:"Extra arguments to pass to llama-cpp-rpc-server"`
+}
+
+func (r *P2P) Run(ctx *cliContext.Context) error {
+	// Extract files from the embedded FS
+	err := assets.ExtractFiles(ctx.BackendAssets, r.BackendAssetsPath)
+	log.Debug().Msgf("Extracting backend assets files to %s", r.BackendAssetsPath)
+	if err != nil {
+		log.Warn().Msgf("Failed extracting backend assets files: %s (might be required for some backends to work properly, like gpt4all)", err)
+	}
+
+	// Check if the token is set
+	// as we always need it.
+	if r.Token == "" {
+		return fmt.Errorf("Token is required")
+	}
+
+	port, err := freeport.GetFreePort()
+	if err != nil {
+		return err
+	}
+
+	address := "127.0.0.1"
+
+	if r.NoRunner {
+		// Let override which port and address to bind if the user
+		// configure the llama-cpp service on its own
+		p := fmt.Sprint(port)
+		if r.RunnerAddress != "" {
+			address = r.RunnerAddress
+		}
+		if r.RunnerPort != "" {
+			p = r.RunnerPort
+		}
+
+		err = p2p.BindLLamaCPPWorker(context.Background(), address, p, r.Token)
+		if err != nil {
+			return err
+		}
+		log.Info().Msgf("You need to start llama-cpp-rpc-server on '%s:%s'", address, p)
+
+		return nil
+	}
+
+	// Start llama.cpp directly from the version we have pre-packaged
+	go func() {
+		for {
+			log.Info().Msgf("Starting llama-cpp-rpc-server on '%s:%d'", address, port)
+			cmd := exec.Command(
+				assets.ResolvePath(
+					r.BackendAssetsPath,
+					"util",
+					"llama-cpp-rpc-server",
+				),
+				append([]string{"--host", address, "--port", fmt.Sprint(port)}, r.ExtraLLamaCPPArgs...)...,
+			)
+
+			cmd.Env = os.Environ()
+
+			cmd.Stderr = os.Stdout
+			cmd.Stdout = os.Stdout
+
+			if err := cmd.Start(); err != nil {
+				log.Error().Err(err).Msg("Failed to start llama-cpp-rpc-server")
+			}
+
+			cmd.Wait()
+		}
+	}()
+
+	err = p2p.BindLLamaCPPWorker(context.Background(), address, fmt.Sprint(port), r.Token)
+	if err != nil {
+		return err
+	}
+
+	for {
+		time.Sleep(1 * time.Second)
+	}
+}