chore: ⬆️ Update ggerganov/llama.cpp to 26a8406ba9198eb6fdd8329fa717555b4f77f05f (#4358)

chore: ⬆️ Update ggerganov/llama.cpp to `26a8406ba9198eb6fdd8329fa717555b4f77f05f` (#4353)

⬆️ Update ggerganov/llama.cpp

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: LocalAI [bot] <139863280+localai-bot@users.noreply.github.com>
Co-authored-by: mudler <2420543+mudler@users.noreply.github.com>

Makefile

@@ -8,7 +8,7 @@ DETECT_LIBS?=true
 # llama.cpp versions
 GOLLAMA_REPO?=https://github.com/go-skynet/go-llama.cpp
 GOLLAMA_VERSION?=2b57a8ae43e4699d3dc5d1496a1ccd42922993be
-CPPLLAMA_VERSION?=cc98896db858df7aa40d0e16a505883ef196a482
+CPPLLAMA_VERSION?=26a8406ba9198eb6fdd8329fa717555b4f77f05f
 
 # whisper.cpp version
 WHISPER_REPO?=https://github.com/ggerganov/whisper.cpp

core/backend/llm.go

@@ -117,12 +117,8 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 			ss := ""
 
 			var partialRune []byte
-			err := inferenceModel.PredictStream(ctx, opts, func(reply *proto.Reply) {
-				msg := reply.GetMessage()
-				partialRune = append(partialRune, msg...)
-
-				tokenUsage.Prompt = int(reply.PromptTokens)
-				tokenUsage.Completion = int(reply.Tokens)
+			err := inferenceModel.PredictStream(ctx, opts, func(chars []byte) {
+				partialRune = append(partialRune, chars...)
 
 				for len(partialRune) > 0 {
 					r, size := utf8.DecodeRune(partialRune)
@@ -136,10 +132,6 @@ func ModelInference(ctx context.Context, s string, messages []schema.Message, im
 
 					partialRune = partialRune[size:]
 				}
-
-				if len(msg) == 0 {
-					tokenCallback("", tokenUsage)
-				}
 			})
 			return LLMResponse{
 				Response: ss,

core/http/endpoints/openai/chat.go

@@ -39,15 +39,11 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
 		responses <- initialMessage
 
 		ComputeChoices(req, s, config, startupOptions, loader, func(s string, c *[]schema.Choice) {}, func(s string, usage backend.TokenUsage) bool {
-			choices := []schema.Choice{}
-			if s != "" {
-				choices = append(choices, schema.Choice{Delta: &schema.Message{Content: &s}, Index: 0})
-			}
 			resp := schema.OpenAIResponse{
 				ID:      id,
 				Created: created,
 				Model:   req.Model, // we have to return what the user sent here, due to OpenAI spec.
-				Choices: choices,
+				Choices: []schema.Choice{{Delta: &schema.Message{Content: &s}, Index: 0}},
 				Object:  "chat.completion.chunk",
 				Usage: schema.OpenAIUsage{
 					PromptTokens:     usage.Prompt,
@@ -469,9 +465,6 @@ func ChatEndpoint(cl *config.BackendConfigLoader, ml *model.ModelLoader, startup
 				toolsCalled := false
 				for ev := range responses {
 					usage = &ev.Usage // Copy a pointer to the latest usage chunk so that the stop message can reference it
-					if len(ev.Choices) == 0 {
-						break
-					}
 					if len(ev.Choices[0].Delta.ToolCalls) > 0 {
 						toolsCalled = true
 					}

pkg/grpc/backend.go

@@ -37,7 +37,7 @@ type Backend interface {
 	Embeddings(ctx context.Context, in *pb.PredictOptions, opts ...grpc.CallOption) (*pb.EmbeddingResult, error)
 	Predict(ctx context.Context, in *pb.PredictOptions, opts ...grpc.CallOption) (*pb.Reply, error)
 	LoadModel(ctx context.Context, in *pb.ModelOptions, opts ...grpc.CallOption) (*pb.Result, error)
-	PredictStream(ctx context.Context, in *pb.PredictOptions, f func(reply *pb.Reply), opts ...grpc.CallOption) error
+	PredictStream(ctx context.Context, in *pb.PredictOptions, f func(s []byte), opts ...grpc.CallOption) error
 	GenerateImage(ctx context.Context, in *pb.GenerateImageRequest, opts ...grpc.CallOption) (*pb.Result, error)
 	TTS(ctx context.Context, in *pb.TTSRequest, opts ...grpc.CallOption) (*pb.Result, error)
 	SoundGeneration(ctx context.Context, in *pb.SoundGenerationRequest, opts ...grpc.CallOption) (*pb.Result, error)

pkg/grpc/client.go

@@ -136,7 +136,7 @@ func (c *Client) LoadModel(ctx context.Context, in *pb.ModelOptions, opts ...grp
 	return client.LoadModel(ctx, in, opts...)
 }
 
-func (c *Client) PredictStream(ctx context.Context, in *pb.PredictOptions, f func(reply *pb.Reply), opts ...grpc.CallOption) error {
+func (c *Client) PredictStream(ctx context.Context, in *pb.PredictOptions, f func(s []byte), opts ...grpc.CallOption) error {
 	if !c.parallel {
 		c.opMutex.Lock()
 		defer c.opMutex.Unlock()
@@ -158,7 +158,7 @@ func (c *Client) PredictStream(ctx context.Context, in *pb.PredictOptions, f fun
 	}
 
 	for {
-		reply, err := stream.Recv()
+		feature, err := stream.Recv()
 		if err == io.EOF {
 			break
 		}
@@ -167,7 +167,7 @@ func (c *Client) PredictStream(ctx context.Context, in *pb.PredictOptions, f fun
 
 			return err
 		}
-		f(reply)
+		f(feature.GetMessage())
 	}
 
 	return nil

pkg/grpc/embed.go

@@ -35,7 +35,7 @@ func (e *embedBackend) LoadModel(ctx context.Context, in *pb.ModelOptions, opts
 	return e.s.LoadModel(ctx, in)
 }
 
-func (e *embedBackend) PredictStream(ctx context.Context, in *pb.PredictOptions, f func(reply *pb.Reply), opts ...grpc.CallOption) error {
+func (e *embedBackend) PredictStream(ctx context.Context, in *pb.PredictOptions, f func(s []byte), opts ...grpc.CallOption) error {
 	bs := &embedBackendServerStream{
 		ctx: ctx,
 		fn:  f,
@@ -97,11 +97,11 @@ func (e *embedBackend) GetTokenMetrics(ctx context.Context, in *pb.MetricsReques
 
 type embedBackendServerStream struct {
 	ctx context.Context
-	fn  func(reply *pb.Reply)
+	fn  func(s []byte)
 }
 
 func (e *embedBackendServerStream) Send(reply *pb.Reply) error {
-	e.fn(reply)
+	e.fn(reply.GetMessage())
 	return nil
 }