-
-
Notifications
You must be signed in to change notification settings - Fork 2k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
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 <[email protected]>
- Loading branch information
Showing
15 changed files
with
1,155 additions
and
43 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -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)"` | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -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:"-"` | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -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"` | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -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") | ||
| } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -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) | ||
| } | ||
| } |
Oops, something went wrong.