File size: 6,271 Bytes
7def60a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
package grpc

import (
	"context"
	"fmt"
	"log"
	"net"

	pb "github.com/mudler/LocalAI/pkg/grpc/proto"
	"google.golang.org/grpc"
)

// A GRPC Server that allows to run LLM inference.
// It is used by the LLMServices to expose the LLM functionalities that are called by the client.
// The GRPC Service is general, trying to encompass all the possible LLM options models.
// It depends on the real implementer then what can be done or not.
//
// The server is implemented as a GRPC service, with the following methods:
// - Predict: to run the inference with options
// - PredictStream: to run the inference with options and stream the results

// server is used to implement helloworld.GreeterServer.
type server struct {
	pb.UnimplementedBackendServer
	llm LLM
}

func (s *server) Health(ctx context.Context, in *pb.HealthMessage) (*pb.Reply, error) {
	return newReply("OK"), nil
}

func (s *server) Embedding(ctx context.Context, in *pb.PredictOptions) (*pb.EmbeddingResult, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	embeds, err := s.llm.Embeddings(in)
	if err != nil {
		return nil, err
	}

	return &pb.EmbeddingResult{Embeddings: embeds}, nil
}

func (s *server) LoadModel(ctx context.Context, in *pb.ModelOptions) (*pb.Result, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	err := s.llm.Load(in)
	if err != nil {
		return &pb.Result{Message: fmt.Sprintf("Error loading model: %s", err.Error()), Success: false}, err
	}
	return &pb.Result{Message: "Loading succeeded", Success: true}, nil
}

func (s *server) Predict(ctx context.Context, in *pb.PredictOptions) (*pb.Reply, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	result, err := s.llm.Predict(in)
	return newReply(result), err
}

func (s *server) GenerateImage(ctx context.Context, in *pb.GenerateImageRequest) (*pb.Result, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	err := s.llm.GenerateImage(in)
	if err != nil {
		return &pb.Result{Message: fmt.Sprintf("Error generating image: %s", err.Error()), Success: false}, err
	}
	return &pb.Result{Message: "Image generated", Success: true}, nil
}

func (s *server) TTS(ctx context.Context, in *pb.TTSRequest) (*pb.Result, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	err := s.llm.TTS(in)
	if err != nil {
		return &pb.Result{Message: fmt.Sprintf("Error generating audio: %s", err.Error()), Success: false}, err
	}
	return &pb.Result{Message: "Audio generated", Success: true}, nil
}

func (s *server) AudioTranscription(ctx context.Context, in *pb.TranscriptRequest) (*pb.TranscriptResult, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	result, err := s.llm.AudioTranscription(in)
	if err != nil {
		return nil, err
	}
	tresult := &pb.TranscriptResult{}
	for _, s := range result.Segments {
		tks := []int32{}
		for _, t := range s.Tokens {
			tks = append(tks, int32(t))
		}
		tresult.Segments = append(tresult.Segments,
			&pb.TranscriptSegment{
				Text:   s.Text,
				Id:     int32(s.Id),
				Start:  int64(s.Start),
				End:    int64(s.End),
				Tokens: tks,
			})
	}

	tresult.Text = result.Text
	return tresult, nil
}

func (s *server) PredictStream(in *pb.PredictOptions, stream pb.Backend_PredictStreamServer) error {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	resultChan := make(chan string)

	done := make(chan bool)
	go func() {
		for result := range resultChan {
			stream.Send(newReply(result))
		}
		done <- true
	}()

	err := s.llm.PredictStream(in, resultChan)
	<-done

	return err
}

func (s *server) TokenizeString(ctx context.Context, in *pb.PredictOptions) (*pb.TokenizationResponse, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	res, err := s.llm.TokenizeString(in)
	if err != nil {
		return nil, err
	}

	castTokens := make([]int32, len(res.Tokens))
	for i, v := range res.Tokens {
		castTokens[i] = int32(v)
	}

	return &pb.TokenizationResponse{
		Length: int32(res.Length),
		Tokens: castTokens,
	}, err
}

func (s *server) Status(ctx context.Context, in *pb.HealthMessage) (*pb.StatusResponse, error) {
	res, err := s.llm.Status()
	if err != nil {
		return nil, err
	}

	return &res, nil
}

func (s *server) StoresSet(ctx context.Context, in *pb.StoresSetOptions) (*pb.Result, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	err := s.llm.StoresSet(in)
	if err != nil {
		return &pb.Result{Message: fmt.Sprintf("Error setting entry: %s", err.Error()), Success: false}, err
	}
	return &pb.Result{Message: "Set key", Success: true}, nil
}

func (s *server) StoresDelete(ctx context.Context, in *pb.StoresDeleteOptions) (*pb.Result, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	err := s.llm.StoresDelete(in)
	if err != nil {
		return &pb.Result{Message: fmt.Sprintf("Error deleting entry: %s", err.Error()), Success: false}, err
	}
	return &pb.Result{Message: "Deleted key", Success: true}, nil
}

func (s *server) StoresGet(ctx context.Context, in *pb.StoresGetOptions) (*pb.StoresGetResult, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	res, err := s.llm.StoresGet(in)
	if err != nil {
		return nil, err
	}
	return &res, nil
}

func (s *server) StoresFind(ctx context.Context, in *pb.StoresFindOptions) (*pb.StoresFindResult, error) {
	if s.llm.Locking() {
		s.llm.Lock()
		defer s.llm.Unlock()
	}
	res, err := s.llm.StoresFind(in)
	if err != nil {
		return nil, err
	}
	return &res, nil
}

func StartServer(address string, model LLM) error {
	lis, err := net.Listen("tcp", address)
	if err != nil {
		return err
	}
	s := grpc.NewServer()
	pb.RegisterBackendServer(s, &server{llm: model})
	log.Printf("gRPC Server listening at %v", lis.Addr())
	if err := s.Serve(lis); err != nil {
		return err
	}

	return nil
}

func RunServer(address string, model LLM) (func() error, error) {
	lis, err := net.Listen("tcp", address)
	if err != nil {
		return nil, err
	}
	s := grpc.NewServer()
	pb.RegisterBackendServer(s, &server{llm: model})
	log.Printf("gRPC Server listening at %v", lis.Addr())
	if err = s.Serve(lis); err != nil {
		return func() error {
			return lis.Close()
		}, err
	}

	return func() error {
		s.GracefulStop()
		return nil
	}, nil
}