feat: Centralized Request Processing middleware

.bruno/LocalAI Test Requests/vad/vad test too few.bru

@@ -0,0 +1,22 @@
+meta {
+  name: vad test too few
+  type: http
+  seq: 1
+}
+
+post {
+  url: {{PROTOCOL}}{{HOST}}:{{PORT}}/vad
+  body: json
+  auth: none
+}
+
+headers {
+  Content-Type: application/json
+}
+
+body:json {
+  {
+      "model": "silero-vad",
+      "audio": []
+  }
+}

aio/cpu/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/entrypoint.sh

@@ -129,7 +129,7 @@ detect_gpu
 detect_gpu_size
 
 PROFILE="${PROFILE:-$GPU_SIZE}" # default to cpu
-export MODELS="${MODELS:-/aio/${PROFILE}/embeddings.yaml,/aio/${PROFILE}/rerank.yaml,/aio/${PROFILE}/text-to-speech.yaml,/aio/${PROFILE}/image-gen.yaml,/aio/${PROFILE}/text-to-text.yaml,/aio/${PROFILE}/speech-to-text.yaml,/aio/${PROFILE}/vision.yaml}"
+export MODELS="${MODELS:-/aio/${PROFILE}/embeddings.yaml,/aio/${PROFILE}/rerank.yaml,/aio/${PROFILE}/text-to-speech.yaml,/aio/${PROFILE}/image-gen.yaml,/aio/${PROFILE}/text-to-text.yaml,/aio/${PROFILE}/speech-to-text.yaml,/aio/${PROFILE}/vad.yaml,/aio/${PROFILE}/vision.yaml}"
 
 check_vars
 

aio/gpu-8g/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

aio/intel/vad.yaml

@@ -0,0 +1,8 @@
+backend: silero-vad
+name: silero-vad
+parameters:
+  model: silero-vad.onnx
+download_files:
+- filename: silero-vad.onnx
+  uri: https://huggingface.co/onnx-community/silero-vad/resolve/main/onnx/model.onnx
+  sha256: a4a068cd6cf1ea8355b84327595838ca748ec29a25bc91fc82e6c299ccdc5808

core/application/startup.go

@@ -145,13 +145,7 @@ func New(opts ...config.AppOption) (*Application, error) {
 
 	if options.LoadToMemory != nil {
 		for _, m := range options.LoadToMemory {
-			cfg, err := application.BackendLoader().LoadBackendConfigFileByName(m, options.ModelPath,
-				config.LoadOptionDebug(options.Debug),
-				config.LoadOptionThreads(options.Threads),
-				config.LoadOptionContextSize(options.ContextSize),
-				config.LoadOptionF16(options.F16),
-				config.ModelPath(options.ModelPath),
-			)
+			cfg, err := application.BackendLoader().LoadBackendConfigFileByNameDefaultOptions(m, options)
 			if err != nil {
 				return nil, err
 			}

core/backend/llm.go

@@ -33,7 +33,7 @@ type TokenUsage struct {
 	TimingTokenGeneration  float64
 }
 
-func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
+func ModelInference(ctx context.Context, s string, messages []schema.Message, images, videos, audios []string, loader *model.ModelLoader, c *config.BackendConfig, o *config.ApplicationConfig, tokenCallback func(string, TokenUsage) bool) (func() (LLMResponse, error), error) {
 	modelFile := c.Model
 
 	// Check if the modelFile exists, if it doesn't try to load it from the gallery
@@ -48,7 +48,7 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 		}
 	}
 
-	opts := ModelOptions(c, o)
+	opts := ModelOptions(*c, o)
 	inferenceModel, err := loader.Load(opts...)
 	if err != nil {
 		return nil, err
@@ -84,7 +84,7 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 
 	// in GRPC, the backend is supposed to answer to 1 single token if stream is not supported
 	fn := func() (LLMResponse, error) {
-		opts := gRPCPredictOpts(c, loader.ModelPath)
+		opts := gRPCPredictOpts(*c, loader.ModelPath)
 		opts.Prompt = s
 		opts.Messages = protoMessages
 		opts.UseTokenizerTemplate = c.TemplateConfig.UseTokenizerTemplate

core/backend/rerank.go

@@ -9,10 +9,10 @@ import (
 	model "github.com/mudler/LocalAI/pkg/model"
 )
 
-func Rerank(modelFile string, request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
-
-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
+func Rerank(request *proto.RerankRequest, loader *model.ModelLoader, appConfig *config.ApplicationConfig, backendConfig config.BackendConfig) (*proto.RerankResult, error) {
+	opts := ModelOptions(backendConfig, appConfig)
 	rerankModel, err := loader.Load(opts...)
+
 	if err != nil {
 		return nil, err
 	}

core/backend/soundgeneration.go

@@ -13,7 +13,6 @@ import (
 )
 
 func SoundGeneration(
-	modelFile string,
 	text string,
 	duration *float32,
 	temperature *float32,
@@ -25,8 +24,9 @@ func SoundGeneration(
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {
 
-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
+	opts := ModelOptions(backendConfig, appConfig)
 	soundGenModel, err := loader.Load(opts...)
+
 	if err != nil {
 		return "", nil, err
 	}
@@ -44,7 +44,7 @@ func SoundGeneration(
 
 	res, err := soundGenModel.SoundGeneration(context.Background(), &proto.SoundGenerationRequest{
 		Text:        text,
-		Model:       modelFile,
+		Model:       backendConfig.Model,
 		Dst:         filePath,
 		Sample:      doSample,
 		Duration:    duration,

core/backend/tokenize.go

@@ -4,18 +4,17 @@ import (
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/pkg/grpc"
-	model "github.com/mudler/LocalAI/pkg/model"
+	"github.com/mudler/LocalAI/pkg/model"
 )
 
 func ModelTokenize(s string, loader *model.ModelLoader, backendConfig config.BackendConfig, appConfig *config.ApplicationConfig) (schema.TokenizeResponse, error) {
 
-	modelFile := backendConfig.Model
-
 	var inferenceModel grpc.Backend
 	var err error
 
-	opts := ModelOptions(backendConfig, appConfig, model.WithModel(modelFile))
+	opts := ModelOptions(backendConfig, appConfig)
 
+	// TODO: looks weird, seems to be a correct merge?
 	if backendConfig.Backend == "" {
 		inferenceModel, err = loader.Load(opts...)
 	} else {

core/backend/transcript.go

@@ -47,7 +47,7 @@ func ModelTranscription(audio, language string, translate bool, ml *model.ModelL
 			tks = append(tks, int(t))
 		}
 		tr.Segments = append(tr.Segments,
-			schema.Segment{
+			schema.TranscriptionSegment{
 				Text:   s.Text,
 				Id:     int(s.Id),
 				Start:  time.Duration(s.Start),

core/backend/tts.go

@@ -14,28 +14,22 @@ import (
 )
 
 func ModelTTS(
-	backend,
 	text,
-	modelFile,
 	voice,
 	language string,
 	loader *model.ModelLoader,
 	appConfig *config.ApplicationConfig,
 	backendConfig config.BackendConfig,
 ) (string, *proto.Result, error) {
-	bb := backend
-	if bb == "" {
-		bb = model.PiperBackend
-	}
-
-	opts := ModelOptions(backendConfig, appConfig, model.WithBackendString(bb), model.WithModel(modelFile))
+	opts := ModelOptions(backendConfig, appConfig, model.WithDefaultBackendString(model.PiperBackend))
 	ttsModel, err := loader.Load(opts...)
+
 	if err != nil {
 		return "", nil, err
 	}
 
 	if ttsModel == nil {
-		return "", nil, fmt.Errorf("could not load piper model")
+		return "", nil, fmt.Errorf("could not load tts model %q", backendConfig.Model)
 	}
 
 	if err := os.MkdirAll(appConfig.AudioDir, 0750); err != nil {
@@ -45,22 +39,21 @@ func ModelTTS(
 	fileName := utils.GenerateUniqueFileName(appConfig.AudioDir, "tts", ".wav")
 	filePath := filepath.Join(appConfig.AudioDir, fileName)
 
-	// If the model file is not empty, we pass it joined with the model path
+	// We join the model name to the model path here. This seems to only be done for TTS and is HIGHLY suspect.
+	// This should be addressed in a follow up PR soon.
+	// Copying it over nearly verbatim, as TTS backends are not functional without this.
 	modelPath := ""
-	if modelFile != "" {
-		// If the model file is not empty, we pass it joined with the model path
-		// Checking first that it exists and is not outside ModelPath
-		// TODO: we should actually first check if the modelFile is looking like
-		// a FS path
-		mp := filepath.Join(loader.ModelPath, modelFile)
-		if _, err := os.Stat(mp); err == nil {
-			if err := utils.VerifyPath(mp, appConfig.ModelPath); err != nil {
-				return "", nil, err
-			}
-			modelPath = mp
-		} else {
-			modelPath = modelFile
+	// Checking first that it exists and is not outside ModelPath
+	// TODO: we should actually first check if the modelFile is looking like
+	// a FS path
+	mp := filepath.Join(loader.ModelPath, backendConfig.Model)
+	if _, err := os.Stat(mp); err == nil {
+		if err := utils.VerifyPath(mp, appConfig.ModelPath); err != nil {
+			return "", nil, err
 		}
+		modelPath = mp
+	} else {
+		modelPath = backendConfig.Model // skip this step if it fails?????
 	}
 
 	res, err := ttsModel.TTS(context.Background(), &proto.TTSRequest{

core/backend/vad.go

@@ -0,0 +1,38 @@
+package backend
+
+import (
+	"context"
+
+	"github.com/mudler/LocalAI/core/config"
+	"github.com/mudler/LocalAI/core/schema"
+	"github.com/mudler/LocalAI/pkg/grpc/proto"
+	"github.com/mudler/LocalAI/pkg/model"
+)
+
+func VAD(request *schema.VADRequest,
+	ctx context.Context,
+	ml *model.ModelLoader,
+	appConfig *config.ApplicationConfig,
+	backendConfig config.BackendConfig) (*schema.VADResponse, error) {
+	opts := ModelOptions(backendConfig, appConfig)
+	vadModel, err := ml.Load(opts...)
+	if err != nil {
+		return nil, err
+	}
+	req := proto.VADRequest{
+		Audio: request.Audio,
+	}
+	resp, err := vadModel.VAD(ctx, &req)
+	if err != nil {
+		return nil, err
+	}
+
+	segments := []schema.VADSegment{}
+	for _, s := range resp.Segments {
+		segments = append(segments, schema.VADSegment{Start: s.Start, End: s.End})
+	}
+
+	return &schema.VADResponse{
+		Segments: segments,
+	}, nil
+}

core/cli/soundgeneration.go

@@ -86,13 +86,14 @@ func (t *SoundGenerationCMD) Run(ctx *cliContext.Context) error {
 	options := config.BackendConfig{}
 	options.SetDefaults()
 	options.Backend = t.Backend
+	options.Model = t.Model
 
 	var inputFile *string
 	if t.InputFile != "" {
 		inputFile = &t.InputFile
 	}
 
-	filePath, _, err := backend.SoundGeneration(t.Model, text,
+	filePath, _, err := backend.SoundGeneration(text,
 		parseToFloat32Ptr(t.Duration), parseToFloat32Ptr(t.Temperature), &t.DoSample,
 		inputFile, parseToInt32Ptr(t.InputFileSampleDivisor), ml, opts, options)
 

core/cli/tts.go

@@ -52,8 +52,10 @@ func (t *TTSCMD) Run(ctx *cliContext.Context) error {
 
 	options := config.BackendConfig{}
 	options.SetDefaults()
+	options.Backend = t.Backend
+	options.Model = t.Model
 
-	filePath, _, err := backend.ModelTTS(t.Backend, text, t.Model, t.Voice, t.Language, ml, opts, options)
+	filePath, _, err := backend.ModelTTS(text, t.Voice, t.Language, ml, opts, options)
 	if err != nil {
 		return err
 	}

core/config/backend_config.go

@@ -436,19 +436,21 @@ func (c *BackendConfig) HasTemplate() bool {
 type BackendConfigUsecases int
 
 const (
-	FLAG_ANY              BackendConfigUsecases = 0b000000000
-	FLAG_CHAT             BackendConfigUsecases = 0b000000001
-	FLAG_COMPLETION       BackendConfigUsecases = 0b000000010
-	FLAG_EDIT             BackendConfigUsecases = 0b000000100
-	FLAG_EMBEDDINGS       BackendConfigUsecases = 0b000001000
-	FLAG_RERANK           BackendConfigUsecases = 0b000010000
-	FLAG_IMAGE            BackendConfigUsecases = 0b000100000
-	FLAG_TRANSCRIPT       BackendConfigUsecases = 0b001000000
-	FLAG_TTS              BackendConfigUsecases = 0b010000000
-	FLAG_SOUND_GENERATION BackendConfigUsecases = 0b100000000
+	FLAG_ANY              BackendConfigUsecases = 0b00000000000
+	FLAG_CHAT             BackendConfigUsecases = 0b00000000001
+	FLAG_COMPLETION       BackendConfigUsecases = 0b00000000010
+	FLAG_EDIT             BackendConfigUsecases = 0b00000000100
+	FLAG_EMBEDDINGS       BackendConfigUsecases = 0b00000001000
+	FLAG_RERANK           BackendConfigUsecases = 0b00000010000
+	FLAG_IMAGE            BackendConfigUsecases = 0b00000100000
+	FLAG_TRANSCRIPT       BackendConfigUsecases = 0b00001000000
+	FLAG_TTS              BackendConfigUsecases = 0b00010000000
+	FLAG_SOUND_GENERATION BackendConfigUsecases = 0b00100000000
+	FLAG_TOKENIZE         BackendConfigUsecases = 0b01000000000
+	FLAG_VAD              BackendConfigUsecases = 0b10000000000
 
 	// Common Subsets
-	FLAG_LLM BackendConfigUsecases = FLAG_CHAT & FLAG_COMPLETION & FLAG_EDIT
+	FLAG_LLM BackendConfigUsecases = FLAG_CHAT | FLAG_COMPLETION | FLAG_EDIT
 )
 
 func GetAllBackendConfigUsecases() map[string]BackendConfigUsecases {
@@ -463,6 +465,8 @@ func GetAllBackendConfigUsecases() map[string]BackendConfigUsecases {
 		"FLAG_TRANSCRIPT":       FLAG_TRANSCRIPT,
 		"FLAG_TTS":              FLAG_TTS,
 		"FLAG_SOUND_GENERATION": FLAG_SOUND_GENERATION,
+		"FLAG_TOKENIZE":         FLAG_TOKENIZE,
+		"FLAG_VAD":              FLAG_VAD,
 		"FLAG_LLM":              FLAG_LLM,
 	}
 }
@@ -548,5 +552,18 @@ func (c *BackendConfig) GuessUsecases(u BackendConfigUsecases) bool {
 		}
 	}
 
+	if (u & FLAG_TOKENIZE) == FLAG_TOKENIZE {
+		tokenizeCapableBackends := []string{"llama.cpp", "rwkv"}
+		if !slices.Contains(tokenizeCapableBackends, c.Backend) {
+			return false
+		}
+	}
+
+	if (u & FLAG_VAD) == FLAG_VAD {
+		if c.Backend != "silero-vad" {
+			return false
+		}
+	}
+
 	return true
 }