diff --git a/Dockerfile b/Dockerfile index f67a1f3eb..18db424b9 100644 --- a/Dockerfile +++ b/Dockerfile @@ -11,15 +11,15 @@ ARG TARGETARCH ARG TARGETVARIANT ENV BUILD_TYPE=${BUILD_TYPE} -ENV EXTERNAL_GRPC_BACKENDS="huggingface-embeddings:/build/extra/grpc/huggingface/huggingface.py" +ENV EXTERNAL_GRPC_BACKENDS="huggingface-embeddings:/build/extra/grpc/huggingface/huggingface.py,autogptq:/build/extra/grpc/autogptq/autogptq.py,bark:/build/extra/grpc/bark/ttsbark.py" ARG GO_TAGS="stablediffusion tts" RUN apt-get update && \ apt-get install -y ca-certificates cmake curl patch pip -# Extras requirements -COPY extra/requirements.txt /build/extra/requirements.txt -RUN pip install -r /build/extra/requirements.txt && rm -rf /build/extra/requirements.txt +# Use the variables in subsequent instructions +RUN echo "Target Architecture: $TARGETARCH" +RUN echo "Target Variant: $TARGETVARIANT" # CuBLAS requirements RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \ @@ -33,6 +33,14 @@ RUN if [ "${BUILD_TYPE}" = "cublas" ]; then \ ; fi ENV PATH /usr/local/cuda/bin:${PATH} +# Extras requirements +COPY extra/requirements.txt /build/extra/requirements.txt + +RUN if [ "${TARGETARCH}" = "amd64" ]; then \ + pip install auto-gptq;\ + fi +RUN pip install -r /build/extra/requirements.txt && rm -rf /build/extra/requirements.txt + WORKDIR /build # OpenBLAS requirements @@ -42,9 +50,6 @@ RUN apt-get install -y libopenblas-dev RUN apt-get install -y libopencv-dev && \ ln -s /usr/include/opencv4/opencv2 /usr/include/opencv2 -# Use the variables in subsequent instructions -RUN echo "Target Architecture: $TARGETARCH" -RUN echo "Target Variant: $TARGETVARIANT" # piper requirements # Use pre-compiled Piper phonemization library (includes onnxruntime) diff --git a/Makefile b/Makefile index 26a90beea..971e6e3c5 100644 --- a/Makefile +++ b/Makefile @@ -335,7 +335,9 @@ protogen-go: pkg/grpc/proto/backend.proto protogen-python: - python -m grpc_tools.protoc -Ipkg/grpc/proto/ --python_out=extra/grpc/huggingface/ --grpc_python_out=extra/grpc/huggingface/ pkg/grpc/proto/backend.proto + python3 -m grpc_tools.protoc -Ipkg/grpc/proto/ --python_out=extra/grpc/huggingface/ --grpc_python_out=extra/grpc/huggingface/ pkg/grpc/proto/backend.proto + python3 -m grpc_tools.protoc -Ipkg/grpc/proto/ --python_out=extra/grpc/autogptq/ --grpc_python_out=extra/grpc/autogptq/ pkg/grpc/proto/backend.proto + python3 -m grpc_tools.protoc -Ipkg/grpc/proto/ --python_out=extra/grpc/bark/ --grpc_python_out=extra/grpc/bark/ pkg/grpc/proto/backend.proto ## GRPC diff --git a/api/backend/embeddings.go b/api/backend/embeddings.go index 53df785b2..c6ebad52e 100644 --- a/api/backend/embeddings.go +++ b/api/backend/embeddings.go @@ -26,7 +26,7 @@ func ModelEmbedding(s string, tokens []int, loader *model.ModelLoader, c config. model.WithLoadGRPCLLMModelOpts(grpcOpts), model.WithThreads(uint32(c.Threads)), model.WithAssetDir(o.AssetsDestination), - model.WithModelFile(modelFile), + model.WithModel(modelFile), model.WithContext(o.Context), } diff --git a/api/backend/image.go b/api/backend/image.go index 9e32d1dba..f3b3a9d75 100644 --- a/api/backend/image.go +++ b/api/backend/image.go @@ -20,7 +20,7 @@ func ImageGeneration(height, width, mode, step, seed int, positive_prompt, negat model.WithAssetDir(o.AssetsDestination), model.WithThreads(uint32(c.Threads)), model.WithContext(o.Context), - model.WithModelFile(c.ImageGenerationAssets), + model.WithModel(c.ImageGenerationAssets), } for k, v := range o.ExternalGRPCBackends { diff --git a/api/backend/llm.go b/api/backend/llm.go index 498c9b09f..804df582b 100644 --- a/api/backend/llm.go +++ b/api/backend/llm.go @@ -27,7 +27,7 @@ func ModelInference(ctx context.Context, s string, loader *model.ModelLoader, c model.WithLoadGRPCLLMModelOpts(grpcOpts), model.WithThreads(uint32(c.Threads)), // some models uses this to allocate threads during startup model.WithAssetDir(o.AssetsDestination), - model.WithModelFile(modelFile), + model.WithModel(modelFile), model.WithContext(o.Context), } diff --git a/api/backend/options.go b/api/backend/options.go index 813ecd7d4..b5df485ec 100644 --- a/api/backend/options.go +++ b/api/backend/options.go @@ -15,23 +15,27 @@ func gRPCModelOpts(c config.Config) *pb.ModelOptions { b = c.Batch } return &pb.ModelOptions{ - ContextSize: int32(c.ContextSize), - Seed: int32(c.Seed), - NBatch: int32(b), - NGQA: c.NGQA, - RMSNormEps: c.RMSNormEps, - F16Memory: c.F16, - MLock: c.MMlock, - RopeFreqBase: c.RopeFreqBase, - RopeFreqScale: c.RopeFreqScale, - NUMA: c.NUMA, - Embeddings: c.Embeddings, - LowVRAM: c.LowVRAM, - NGPULayers: int32(c.NGPULayers), - MMap: c.MMap, - MainGPU: c.MainGPU, - Threads: int32(c.Threads), - TensorSplit: c.TensorSplit, + ContextSize: int32(c.ContextSize), + Seed: int32(c.Seed), + NBatch: int32(b), + NGQA: c.NGQA, + ModelBaseName: c.ModelBaseName, + UseFastTokenizer: c.UseFastTokenizer, + Device: c.Device, + UseTriton: c.Triton, + RMSNormEps: c.RMSNormEps, + F16Memory: c.F16, + MLock: c.MMlock, + RopeFreqBase: c.RopeFreqBase, + RopeFreqScale: c.RopeFreqScale, + NUMA: c.NUMA, + Embeddings: c.Embeddings, + LowVRAM: c.LowVRAM, + NGPULayers: int32(c.NGPULayers), + MMap: c.MMap, + MainGPU: c.MainGPU, + Threads: int32(c.Threads), + TensorSplit: c.TensorSplit, } } diff --git a/api/backend/transcript.go b/api/backend/transcript.go index b2f25012c..add16f4cf 100644 --- a/api/backend/transcript.go +++ b/api/backend/transcript.go @@ -15,7 +15,7 @@ import ( func ModelTranscription(audio, language string, loader *model.ModelLoader, c config.Config, o *options.Option) (*api.Result, error) { opts := []model.Option{ model.WithBackendString(model.WhisperBackend), - model.WithModelFile(c.Model), + model.WithModel(c.Model), model.WithContext(o.Context), model.WithThreads(uint32(c.Threads)), model.WithAssetDir(o.AssetsDestination), diff --git a/api/backend/tts.go b/api/backend/tts.go index ac491e250..411fc278d 100644 --- a/api/backend/tts.go +++ b/api/backend/tts.go @@ -28,10 +28,14 @@ func generateUniqueFileName(dir, baseName, ext string) string { } } -func ModelTTS(text, modelFile string, loader *model.ModelLoader, o *options.Option) (string, *proto.Result, error) { +func ModelTTS(backend, text, modelFile string, loader *model.ModelLoader, o *options.Option) (string, *proto.Result, error) { + bb := backend + if bb == "" { + bb = model.PiperBackend + } opts := []model.Option{ - model.WithBackendString(model.PiperBackend), - model.WithModelFile(modelFile), + model.WithBackendString(bb), + model.WithModel(modelFile), model.WithContext(o.Context), model.WithAssetDir(o.AssetsDestination), } @@ -56,10 +60,13 @@ func ModelTTS(text, modelFile string, loader *model.ModelLoader, o *options.Opti fileName := generateUniqueFileName(o.AudioDir, "piper", ".wav") filePath := filepath.Join(o.AudioDir, fileName) - modelPath := filepath.Join(o.Loader.ModelPath, modelFile) - - if err := utils.VerifyPath(modelPath, o.Loader.ModelPath); err != nil { - return "", nil, err + // If the model file is not empty, we pass it joined with the model path + modelPath := "" + if modelFile != "" { + modelPath = filepath.Join(o.Loader.ModelPath, modelFile) + if err := utils.VerifyPath(modelPath, o.Loader.ModelPath); err != nil { + return "", nil, err + } } res, err := piperModel.TTS(context.Background(), &proto.TTSRequest{ diff --git a/api/config/config.go b/api/config/config.go index 757827155..47ce34087 100644 --- a/api/config/config.go +++ b/api/config/config.go @@ -54,6 +54,12 @@ type Config struct { RMSNormEps float32 `yaml:"rms_norm_eps"` NGQA int32 `yaml:"ngqa"` + + // AutoGPTQ + ModelBaseName string `yaml:"model_base_name"` + Device string `yaml:"device"` + Triton bool `yaml:"triton"` + UseFastTokenizer bool `yaml:"use_fast_tokenizer"` } type Functions struct { diff --git a/api/config/prediction.go b/api/config/prediction.go index 85621fa99..9cd1d2b12 100644 --- a/api/config/prediction.go +++ b/api/config/prediction.go @@ -39,4 +39,6 @@ type PredictionOptions struct { RopeFreqBase float32 `json:"rope_freq_base" yaml:"rope_freq_base"` RopeFreqScale float32 `json:"rope_freq_scale" yaml:"rope_freq_scale"` NegativePromptScale float32 `json:"negative_prompt_scale" yaml:"negative_prompt_scale"` + // AutoGPTQ + UseFastTokenizer bool `json:"use_fast_tokenizer" yaml:"use_fast_tokenizer"` } diff --git a/api/localai/localai.go b/api/localai/localai.go index 49f778057..c9aee2ae5 100644 --- a/api/localai/localai.go +++ b/api/localai/localai.go @@ -9,8 +9,9 @@ import ( ) type TTSRequest struct { - Model string `json:"model" yaml:"model"` - Input string `json:"input" yaml:"input"` + Model string `json:"model" yaml:"model"` + Input string `json:"input" yaml:"input"` + Backend string `json:"backend" yaml:"backend"` } func TTSEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) error { @@ -22,7 +23,7 @@ func TTSEndpoint(cm *config.ConfigLoader, o *options.Option) func(c *fiber.Ctx) return err } - filePath, _, err := backend.ModelTTS(input.Input, input.Model, o.Loader, o) + filePath, _, err := backend.ModelTTS(input.Backend, input.Input, input.Model, o.Loader, o) if err != nil { return err } diff --git a/api/openai/api.go b/api/openai/api.go index ac757f477..2e3f154c7 100644 --- a/api/openai/api.go +++ b/api/openai/api.go @@ -2,6 +2,7 @@ package openai import ( "context" + config "github.com/go-skynet/LocalAI/api/config" "github.com/go-skynet/LocalAI/pkg/grammar" @@ -106,4 +107,9 @@ type OpenAIRequest struct { Grammar string `json:"grammar" yaml:"grammar"` JSONFunctionGrammarObject *grammar.JSONFunctionStructure `json:"grammar_json_functions" yaml:"grammar_json_functions"` + + Backend string `json:"backend" yaml:"backend"` + + // AutoGPTQ + ModelBaseName string `json:"model_base_name" yaml:"model_base_name"` } diff --git a/api/openai/request.go b/api/openai/request.go index 5aa4d46a0..2938757bd 100644 --- a/api/openai/request.go +++ b/api/openai/request.go @@ -71,10 +71,22 @@ func updateConfig(config *config.Config, input *OpenAIRequest) { config.TopP = input.TopP } + if input.Backend != "" { + config.Backend = input.Backend + } + + if input.ModelBaseName != "" { + config.ModelBaseName = input.ModelBaseName + } + if input.NegativePromptScale != 0 { config.NegativePromptScale = input.NegativePromptScale } + if input.UseFastTokenizer { + config.UseFastTokenizer = input.UseFastTokenizer + } + if input.NegativePrompt != "" { config.NegativePrompt = input.NegativePrompt } diff --git a/extra/grpc/autogptq/autogptq.py b/extra/grpc/autogptq/autogptq.py new file mode 100755 index 000000000..7d8a45fca --- /dev/null +++ b/extra/grpc/autogptq/autogptq.py @@ -0,0 +1,109 @@ +#!/usr/bin/env python3 +import grpc +from concurrent import futures +import time +import backend_pb2 +import backend_pb2_grpc +import argparse +import signal +import sys +import os +from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig +from pathlib import Path +from transformers import AutoTokenizer +from transformers import TextGenerationPipeline + +_ONE_DAY_IN_SECONDS = 60 * 60 * 24 + +# Implement the BackendServicer class with the service methods +class BackendServicer(backend_pb2_grpc.BackendServicer): + def Health(self, request, context): + return backend_pb2.Reply(message=bytes("OK", 'utf-8')) + def LoadModel(self, request, context): + try: + device = "cuda:0" + if request.Device != "": + device = request.Device + + tokenizer = AutoTokenizer.from_pretrained(request.Model, use_fast=request.UseFastTokenizer) + + model = AutoGPTQForCausalLM.from_quantized(request.Model, + model_basename=request.ModelBaseName, + use_safetensors=True, + trust_remote_code=True, + device=device, + use_triton=request.UseTriton, + quantize_config=None) + + self.model = model + self.tokenizer = tokenizer + except Exception as err: + return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}") + return backend_pb2.Result(message="Model loaded successfully", success=True) + + def Predict(self, request, context): + penalty = 1.0 + if request.Penalty != 0.0: + penalty = request.Penalty + tokens = 512 + if request.Tokens != 0: + tokens = request.Tokens + top_p = 0.95 + if request.TopP != 0.0: + top_p = request.TopP + + # Implement Predict RPC + pipeline = TextGenerationPipeline( + model=self.model, + tokenizer=self.tokenizer, + max_new_tokens=tokens, + temperature=request.Temperature, + top_p=top_p, + repetition_penalty=penalty, + ) + t = pipeline(request.Prompt)[0]["generated_text"] + # Remove prompt from response if present + if request.Prompt in t: + t = t.replace(request.Prompt, "") + + return backend_pb2.Result(message=bytes(t, encoding='utf-8')) + + def PredictStream(self, request, context): + # Implement PredictStream RPC + #for reply in some_data_generator(): + # yield reply + # Not implemented yet + return self.Predict(request, context) + + +def serve(address): + server = grpc.server(futures.ThreadPoolExecutor(max_workers=10)) + backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server) + server.add_insecure_port(address) + server.start() + print("Server started. Listening on: " + address, file=sys.stderr) + + # Define the signal handler function + def signal_handler(sig, frame): + print("Received termination signal. Shutting down...") + server.stop(0) + sys.exit(0) + + # Set the signal handlers for SIGINT and SIGTERM + signal.signal(signal.SIGINT, signal_handler) + signal.signal(signal.SIGTERM, signal_handler) + + try: + while True: + time.sleep(_ONE_DAY_IN_SECONDS) + except KeyboardInterrupt: + server.stop(0) + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description="Run the gRPC server.") + parser.add_argument( + "--addr", default="localhost:50051", help="The address to bind the server to." + ) + args = parser.parse_args() + + serve(args.addr) \ No newline at end of file diff --git a/extra/grpc/autogptq/backend_pb2.py b/extra/grpc/autogptq/backend_pb2.py new file mode 100644 index 000000000..e8d2d30c8 --- /dev/null +++ b/extra/grpc/autogptq/backend_pb2.py @@ -0,0 +1,49 @@ +# -*- coding: utf-8 -*- +# Generated by the protocol buffer compiler. DO NOT EDIT! +# source: backend.proto +"""Generated protocol buffer code.""" +from google.protobuf import descriptor as _descriptor +from google.protobuf import descriptor_pool as _descriptor_pool +from google.protobuf import symbol_database as _symbol_database +from google.protobuf.internal import builder as _builder +# @@protoc_insertion_point(imports) + +_sym_db = _symbol_database.Default() + + + + +DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\rbackend.proto\x12\x07\x62\x61\x63kend\"\x0f\n\rHealthMessage\"\x86\x06\n\x0ePredictOptions\x12\x0e\n\x06Prompt\x18\x01 \x01(\t\x12\x0c\n\x04Seed\x18\x02 \x01(\x05\x12\x0f\n\x07Threads\x18\x03 \x01(\x05\x12\x0e\n\x06Tokens\x18\x04 \x01(\x05\x12\x0c\n\x04TopK\x18\x05 \x01(\x05\x12\x0e\n\x06Repeat\x18\x06 \x01(\x05\x12\r\n\x05\x42\x61tch\x18\x07 \x01(\x05\x12\r\n\x05NKeep\x18\x08 \x01(\x05\x12\x13\n\x0bTemperature\x18\t \x01(\x02\x12\x0f\n\x07Penalty\x18\n \x01(\x02\x12\r\n\x05\x46\x31\x36KV\x18\x0b \x01(\x08\x12\x11\n\tDebugMode\x18\x0c \x01(\x08\x12\x13\n\x0bStopPrompts\x18\r \x03(\t\x12\x11\n\tIgnoreEOS\x18\x0e \x01(\x08\x12\x19\n\x11TailFreeSamplingZ\x18\x0f \x01(\x02\x12\x10\n\x08TypicalP\x18\x10 \x01(\x02\x12\x18\n\x10\x46requencyPenalty\x18\x11 \x01(\x02\x12\x17\n\x0fPresencePenalty\x18\x12 \x01(\x02\x12\x10\n\x08Mirostat\x18\x13 \x01(\x05\x12\x13\n\x0bMirostatETA\x18\x14 \x01(\x02\x12\x13\n\x0bMirostatTAU\x18\x15 \x01(\x02\x12\x12\n\nPenalizeNL\x18\x16 \x01(\x08\x12\x11\n\tLogitBias\x18\x17 \x01(\t\x12\r\n\x05MLock\x18\x19 \x01(\x08\x12\x0c\n\x04MMap\x18\x1a \x01(\x08\x12\x16\n\x0ePromptCacheAll\x18\x1b \x01(\x08\x12\x15\n\rPromptCacheRO\x18\x1c \x01(\x08\x12\x0f\n\x07Grammar\x18\x1d \x01(\t\x12\x0f\n\x07MainGPU\x18\x1e \x01(\t\x12\x13\n\x0bTensorSplit\x18\x1f \x01(\t\x12\x0c\n\x04TopP\x18 \x01(\x02\x12\x17\n\x0fPromptCachePath\x18! \x01(\t\x12\r\n\x05\x44\x65\x62ug\x18\" \x01(\x08\x12\x17\n\x0f\x45mbeddingTokens\x18# \x03(\x05\x12\x12\n\nEmbeddings\x18$ \x01(\t\x12\x14\n\x0cRopeFreqBase\x18% \x01(\x02\x12\x15\n\rRopeFreqScale\x18& \x01(\x02\x12\x1b\n\x13NegativePromptScale\x18\' \x01(\x02\x12\x16\n\x0eNegativePrompt\x18( \x01(\t\"\x18\n\x05Reply\x12\x0f\n\x07message\x18\x01 \x01(\x0c\"\xe2\x03\n\x0cModelOptions\x12\r\n\x05Model\x18\x01 \x01(\t\x12\x13\n\x0b\x43ontextSize\x18\x02 \x01(\x05\x12\x0c\n\x04Seed\x18\x03 \x01(\x05\x12\x0e\n\x06NBatch\x18\x04 \x01(\x05\x12\x11\n\tF16Memory\x18\x05 \x01(\x08\x12\r\n\x05MLock\x18\x06 \x01(\x08\x12\x0c\n\x04MMap\x18\x07 \x01(\x08\x12\x11\n\tVocabOnly\x18\x08 \x01(\x08\x12\x0f\n\x07LowVRAM\x18\t \x01(\x08\x12\x12\n\nEmbeddings\x18\n \x01(\x08\x12\x0c\n\x04NUMA\x18\x0b \x01(\x08\x12\x12\n\nNGPULayers\x18\x0c \x01(\x05\x12\x0f\n\x07MainGPU\x18\r \x01(\t\x12\x13\n\x0bTensorSplit\x18\x0e \x01(\t\x12\x0f\n\x07Threads\x18\x0f \x01(\x05\x12\x19\n\x11LibrarySearchPath\x18\x10 \x01(\t\x12\x14\n\x0cRopeFreqBase\x18\x11 \x01(\x02\x12\x15\n\rRopeFreqScale\x18\x12 \x01(\x02\x12\x12\n\nRMSNormEps\x18\x13 \x01(\x02\x12\x0c\n\x04NGQA\x18\x14 \x01(\x05\x12\x11\n\tModelFile\x18\x15 \x01(\t\x12\x0e\n\x06\x44\x65vice\x18\x16 \x01(\t\x12\x11\n\tUseTriton\x18\x17 \x01(\x08\x12\x15\n\rModelBaseName\x18\x18 \x01(\t\x12\x18\n\x10UseFastTokenizer\x18\x19 \x01(\x08\"*\n\x06Result\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0f\n\x07success\x18\x02 \x01(\x08\"%\n\x0f\x45mbeddingResult\x12\x12\n\nembeddings\x18\x01 \x03(\x02\"C\n\x11TranscriptRequest\x12\x0b\n\x03\x64st\x18\x02 \x01(\t\x12\x10\n\x08language\x18\x03 \x01(\t\x12\x0f\n\x07threads\x18\x04 \x01(\r\"N\n\x10TranscriptResult\x12,\n\x08segments\x18\x01 \x03(\x0b\x32\x1a.backend.TranscriptSegment\x12\x0c\n\x04text\x18\x02 \x01(\t\"Y\n\x11TranscriptSegment\x12\n\n\x02id\x18\x01 \x01(\x05\x12\r\n\x05start\x18\x02 \x01(\x03\x12\x0b\n\x03\x65nd\x18\x03 \x01(\x03\x12\x0c\n\x04text\x18\x04 \x01(\t\x12\x0e\n\x06tokens\x18\x05 \x03(\x05\"\x9e\x01\n\x14GenerateImageRequest\x12\x0e\n\x06height\x18\x01 \x01(\x05\x12\r\n\x05width\x18\x02 \x01(\x05\x12\x0c\n\x04mode\x18\x03 \x01(\x05\x12\x0c\n\x04step\x18\x04 \x01(\x05\x12\x0c\n\x04seed\x18\x05 \x01(\x05\x12\x17\n\x0fpositive_prompt\x18\x06 \x01(\t\x12\x17\n\x0fnegative_prompt\x18\x07 \x01(\t\x12\x0b\n\x03\x64st\x18\x08 \x01(\t\"6\n\nTTSRequest\x12\x0c\n\x04text\x18\x01 \x01(\t\x12\r\n\x05model\x18\x02 \x01(\t\x12\x0b\n\x03\x64st\x18\x03 \x01(\t2\xeb\x03\n\x07\x42\x61\x63kend\x12\x32\n\x06Health\x12\x16.backend.HealthMessage\x1a\x0e.backend.Reply\"\x00\x12\x34\n\x07Predict\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x12\x35\n\tLoadModel\x12\x15.backend.ModelOptions\x1a\x0f.backend.Result\"\x00\x12<\n\rPredictStream\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x30\x01\x12@\n\tEmbedding\x12\x17.backend.PredictOptions\x1a\x18.backend.EmbeddingResult\"\x00\x12\x41\n\rGenerateImage\x12\x1d.backend.GenerateImageRequest\x1a\x0f.backend.Result\"\x00\x12M\n\x12\x41udioTranscription\x12\x1a.backend.TranscriptRequest\x1a\x19.backend.TranscriptResult\"\x00\x12-\n\x03TTS\x12\x13.backend.TTSRequest\x1a\x0f.backend.Result\"\x00\x42Z\n\x19io.skynet.localai.backendB\x0eLocalAIBackendP\x01Z+github.com/go-skynet/LocalAI/pkg/grpc/protob\x06proto3') + +_globals = globals() +_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals) +_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'backend_pb2', _globals) +if _descriptor._USE_C_DESCRIPTORS == False: + + DESCRIPTOR._options = None + DESCRIPTOR._serialized_options = b'\n\031io.skynet.localai.backendB\016LocalAIBackendP\001Z+github.com/go-skynet/LocalAI/pkg/grpc/proto' + _globals['_HEALTHMESSAGE']._serialized_start=26 + _globals['_HEALTHMESSAGE']._serialized_end=41 + _globals['_PREDICTOPTIONS']._serialized_start=44 + _globals['_PREDICTOPTIONS']._serialized_end=818 + _globals['_REPLY']._serialized_start=820 + _globals['_REPLY']._serialized_end=844 + _globals['_MODELOPTIONS']._serialized_start=847 + _globals['_MODELOPTIONS']._serialized_end=1329 + _globals['_RESULT']._serialized_start=1331 + _globals['_RESULT']._serialized_end=1373 + _globals['_EMBEDDINGRESULT']._serialized_start=1375 + _globals['_EMBEDDINGRESULT']._serialized_end=1412 + _globals['_TRANSCRIPTREQUEST']._serialized_start=1414 + _globals['_TRANSCRIPTREQUEST']._serialized_end=1481 + _globals['_TRANSCRIPTRESULT']._serialized_start=1483 + _globals['_TRANSCRIPTRESULT']._serialized_end=1561 + _globals['_TRANSCRIPTSEGMENT']._serialized_start=1563 + _globals['_TRANSCRIPTSEGMENT']._serialized_end=1652 + _globals['_GENERATEIMAGEREQUEST']._serialized_start=1655 + _globals['_GENERATEIMAGEREQUEST']._serialized_end=1813 + _globals['_TTSREQUEST']._serialized_start=1815 + _globals['_TTSREQUEST']._serialized_end=1869 + _globals['_BACKEND']._serialized_start=1872 + _globals['_BACKEND']._serialized_end=2363 +# @@protoc_insertion_point(module_scope) diff --git a/extra/grpc/autogptq/backend_pb2_grpc.py b/extra/grpc/autogptq/backend_pb2_grpc.py new file mode 100644 index 000000000..301c0729a --- /dev/null +++ b/extra/grpc/autogptq/backend_pb2_grpc.py @@ -0,0 +1,297 @@ +# Generated by the gRPC Python protocol compiler plugin. DO NOT EDIT! +"""Client and server classes corresponding to protobuf-defined services.""" +import grpc + +import backend_pb2 as backend__pb2 + + +class BackendStub(object): + """Missing associated documentation comment in .proto file.""" + + def __init__(self, channel): + """Constructor. + + Args: + channel: A grpc.Channel. + """ + self.Health = channel.unary_unary( + '/backend.Backend/Health', + request_serializer=backend__pb2.HealthMessage.SerializeToString, + response_deserializer=backend__pb2.Reply.FromString, + ) + self.Predict = channel.unary_unary( + '/backend.Backend/Predict', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.Reply.FromString, + ) + self.LoadModel = channel.unary_unary( + '/backend.Backend/LoadModel', + request_serializer=backend__pb2.ModelOptions.SerializeToString, + response_deserializer=backend__pb2.Result.FromString, + ) + self.PredictStream = channel.unary_stream( + '/backend.Backend/PredictStream', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.Reply.FromString, + ) + self.Embedding = channel.unary_unary( + '/backend.Backend/Embedding', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.EmbeddingResult.FromString, + ) + self.GenerateImage = channel.unary_unary( + '/backend.Backend/GenerateImage', + request_serializer=backend__pb2.GenerateImageRequest.SerializeToString, + response_deserializer=backend__pb2.Result.FromString, + ) + self.AudioTranscription = channel.unary_unary( + '/backend.Backend/AudioTranscription', + request_serializer=backend__pb2.TranscriptRequest.SerializeToString, + response_deserializer=backend__pb2.TranscriptResult.FromString, + ) + self.TTS = channel.unary_unary( + '/backend.Backend/TTS', + request_serializer=backend__pb2.TTSRequest.SerializeToString, + response_deserializer=backend__pb2.Result.FromString, + ) + + +class BackendServicer(object): + """Missing associated documentation comment in .proto file.""" + + def Health(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def Predict(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def LoadModel(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def PredictStream(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def Embedding(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def GenerateImage(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def AudioTranscription(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def TTS(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + +def add_BackendServicer_to_server(servicer, server): + rpc_method_handlers = { + 'Health': grpc.unary_unary_rpc_method_handler( + servicer.Health, + request_deserializer=backend__pb2.HealthMessage.FromString, + response_serializer=backend__pb2.Reply.SerializeToString, + ), + 'Predict': grpc.unary_unary_rpc_method_handler( + servicer.Predict, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.Reply.SerializeToString, + ), + 'LoadModel': grpc.unary_unary_rpc_method_handler( + servicer.LoadModel, + request_deserializer=backend__pb2.ModelOptions.FromString, + response_serializer=backend__pb2.Result.SerializeToString, + ), + 'PredictStream': grpc.unary_stream_rpc_method_handler( + servicer.PredictStream, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.Reply.SerializeToString, + ), + 'Embedding': grpc.unary_unary_rpc_method_handler( + servicer.Embedding, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.EmbeddingResult.SerializeToString, + ), + 'GenerateImage': grpc.unary_unary_rpc_method_handler( + servicer.GenerateImage, + request_deserializer=backend__pb2.GenerateImageRequest.FromString, + response_serializer=backend__pb2.Result.SerializeToString, + ), + 'AudioTranscription': grpc.unary_unary_rpc_method_handler( + servicer.AudioTranscription, + request_deserializer=backend__pb2.TranscriptRequest.FromString, + response_serializer=backend__pb2.TranscriptResult.SerializeToString, + ), + 'TTS': grpc.unary_unary_rpc_method_handler( + servicer.TTS, + request_deserializer=backend__pb2.TTSRequest.FromString, + response_serializer=backend__pb2.Result.SerializeToString, + ), + } + generic_handler = grpc.method_handlers_generic_handler( + 'backend.Backend', rpc_method_handlers) + server.add_generic_rpc_handlers((generic_handler,)) + + + # This class is part of an EXPERIMENTAL API. +class Backend(object): + """Missing associated documentation comment in .proto file.""" + + @staticmethod + def Health(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/Health', + backend__pb2.HealthMessage.SerializeToString, + backend__pb2.Reply.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def Predict(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/Predict', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.Reply.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def LoadModel(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/LoadModel', + backend__pb2.ModelOptions.SerializeToString, + backend__pb2.Result.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def PredictStream(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_stream(request, target, '/backend.Backend/PredictStream', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.Reply.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def Embedding(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/Embedding', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.EmbeddingResult.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def GenerateImage(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/GenerateImage', + backend__pb2.GenerateImageRequest.SerializeToString, + backend__pb2.Result.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def AudioTranscription(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/AudioTranscription', + backend__pb2.TranscriptRequest.SerializeToString, + backend__pb2.TranscriptResult.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def TTS(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/TTS', + backend__pb2.TTSRequest.SerializeToString, + backend__pb2.Result.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) diff --git a/extra/grpc/bark/backend_pb2.py b/extra/grpc/bark/backend_pb2.py new file mode 100644 index 000000000..e8d2d30c8 --- /dev/null +++ b/extra/grpc/bark/backend_pb2.py @@ -0,0 +1,49 @@ +# -*- coding: utf-8 -*- +# Generated by the protocol buffer compiler. DO NOT EDIT! +# source: backend.proto +"""Generated protocol buffer code.""" +from google.protobuf import descriptor as _descriptor +from google.protobuf import descriptor_pool as _descriptor_pool +from google.protobuf import symbol_database as _symbol_database +from google.protobuf.internal import builder as _builder +# @@protoc_insertion_point(imports) + +_sym_db = _symbol_database.Default() + + + + +DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\rbackend.proto\x12\x07\x62\x61\x63kend\"\x0f\n\rHealthMessage\"\x86\x06\n\x0ePredictOptions\x12\x0e\n\x06Prompt\x18\x01 \x01(\t\x12\x0c\n\x04Seed\x18\x02 \x01(\x05\x12\x0f\n\x07Threads\x18\x03 \x01(\x05\x12\x0e\n\x06Tokens\x18\x04 \x01(\x05\x12\x0c\n\x04TopK\x18\x05 \x01(\x05\x12\x0e\n\x06Repeat\x18\x06 \x01(\x05\x12\r\n\x05\x42\x61tch\x18\x07 \x01(\x05\x12\r\n\x05NKeep\x18\x08 \x01(\x05\x12\x13\n\x0bTemperature\x18\t \x01(\x02\x12\x0f\n\x07Penalty\x18\n \x01(\x02\x12\r\n\x05\x46\x31\x36KV\x18\x0b \x01(\x08\x12\x11\n\tDebugMode\x18\x0c \x01(\x08\x12\x13\n\x0bStopPrompts\x18\r \x03(\t\x12\x11\n\tIgnoreEOS\x18\x0e \x01(\x08\x12\x19\n\x11TailFreeSamplingZ\x18\x0f \x01(\x02\x12\x10\n\x08TypicalP\x18\x10 \x01(\x02\x12\x18\n\x10\x46requencyPenalty\x18\x11 \x01(\x02\x12\x17\n\x0fPresencePenalty\x18\x12 \x01(\x02\x12\x10\n\x08Mirostat\x18\x13 \x01(\x05\x12\x13\n\x0bMirostatETA\x18\x14 \x01(\x02\x12\x13\n\x0bMirostatTAU\x18\x15 \x01(\x02\x12\x12\n\nPenalizeNL\x18\x16 \x01(\x08\x12\x11\n\tLogitBias\x18\x17 \x01(\t\x12\r\n\x05MLock\x18\x19 \x01(\x08\x12\x0c\n\x04MMap\x18\x1a \x01(\x08\x12\x16\n\x0ePromptCacheAll\x18\x1b \x01(\x08\x12\x15\n\rPromptCacheRO\x18\x1c \x01(\x08\x12\x0f\n\x07Grammar\x18\x1d \x01(\t\x12\x0f\n\x07MainGPU\x18\x1e \x01(\t\x12\x13\n\x0bTensorSplit\x18\x1f \x01(\t\x12\x0c\n\x04TopP\x18 \x01(\x02\x12\x17\n\x0fPromptCachePath\x18! \x01(\t\x12\r\n\x05\x44\x65\x62ug\x18\" \x01(\x08\x12\x17\n\x0f\x45mbeddingTokens\x18# \x03(\x05\x12\x12\n\nEmbeddings\x18$ \x01(\t\x12\x14\n\x0cRopeFreqBase\x18% \x01(\x02\x12\x15\n\rRopeFreqScale\x18& \x01(\x02\x12\x1b\n\x13NegativePromptScale\x18\' \x01(\x02\x12\x16\n\x0eNegativePrompt\x18( \x01(\t\"\x18\n\x05Reply\x12\x0f\n\x07message\x18\x01 \x01(\x0c\"\xe2\x03\n\x0cModelOptions\x12\r\n\x05Model\x18\x01 \x01(\t\x12\x13\n\x0b\x43ontextSize\x18\x02 \x01(\x05\x12\x0c\n\x04Seed\x18\x03 \x01(\x05\x12\x0e\n\x06NBatch\x18\x04 \x01(\x05\x12\x11\n\tF16Memory\x18\x05 \x01(\x08\x12\r\n\x05MLock\x18\x06 \x01(\x08\x12\x0c\n\x04MMap\x18\x07 \x01(\x08\x12\x11\n\tVocabOnly\x18\x08 \x01(\x08\x12\x0f\n\x07LowVRAM\x18\t \x01(\x08\x12\x12\n\nEmbeddings\x18\n \x01(\x08\x12\x0c\n\x04NUMA\x18\x0b \x01(\x08\x12\x12\n\nNGPULayers\x18\x0c \x01(\x05\x12\x0f\n\x07MainGPU\x18\r \x01(\t\x12\x13\n\x0bTensorSplit\x18\x0e \x01(\t\x12\x0f\n\x07Threads\x18\x0f \x01(\x05\x12\x19\n\x11LibrarySearchPath\x18\x10 \x01(\t\x12\x14\n\x0cRopeFreqBase\x18\x11 \x01(\x02\x12\x15\n\rRopeFreqScale\x18\x12 \x01(\x02\x12\x12\n\nRMSNormEps\x18\x13 \x01(\x02\x12\x0c\n\x04NGQA\x18\x14 \x01(\x05\x12\x11\n\tModelFile\x18\x15 \x01(\t\x12\x0e\n\x06\x44\x65vice\x18\x16 \x01(\t\x12\x11\n\tUseTriton\x18\x17 \x01(\x08\x12\x15\n\rModelBaseName\x18\x18 \x01(\t\x12\x18\n\x10UseFastTokenizer\x18\x19 \x01(\x08\"*\n\x06Result\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0f\n\x07success\x18\x02 \x01(\x08\"%\n\x0f\x45mbeddingResult\x12\x12\n\nembeddings\x18\x01 \x03(\x02\"C\n\x11TranscriptRequest\x12\x0b\n\x03\x64st\x18\x02 \x01(\t\x12\x10\n\x08language\x18\x03 \x01(\t\x12\x0f\n\x07threads\x18\x04 \x01(\r\"N\n\x10TranscriptResult\x12,\n\x08segments\x18\x01 \x03(\x0b\x32\x1a.backend.TranscriptSegment\x12\x0c\n\x04text\x18\x02 \x01(\t\"Y\n\x11TranscriptSegment\x12\n\n\x02id\x18\x01 \x01(\x05\x12\r\n\x05start\x18\x02 \x01(\x03\x12\x0b\n\x03\x65nd\x18\x03 \x01(\x03\x12\x0c\n\x04text\x18\x04 \x01(\t\x12\x0e\n\x06tokens\x18\x05 \x03(\x05\"\x9e\x01\n\x14GenerateImageRequest\x12\x0e\n\x06height\x18\x01 \x01(\x05\x12\r\n\x05width\x18\x02 \x01(\x05\x12\x0c\n\x04mode\x18\x03 \x01(\x05\x12\x0c\n\x04step\x18\x04 \x01(\x05\x12\x0c\n\x04seed\x18\x05 \x01(\x05\x12\x17\n\x0fpositive_prompt\x18\x06 \x01(\t\x12\x17\n\x0fnegative_prompt\x18\x07 \x01(\t\x12\x0b\n\x03\x64st\x18\x08 \x01(\t\"6\n\nTTSRequest\x12\x0c\n\x04text\x18\x01 \x01(\t\x12\r\n\x05model\x18\x02 \x01(\t\x12\x0b\n\x03\x64st\x18\x03 \x01(\t2\xeb\x03\n\x07\x42\x61\x63kend\x12\x32\n\x06Health\x12\x16.backend.HealthMessage\x1a\x0e.backend.Reply\"\x00\x12\x34\n\x07Predict\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x12\x35\n\tLoadModel\x12\x15.backend.ModelOptions\x1a\x0f.backend.Result\"\x00\x12<\n\rPredictStream\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x30\x01\x12@\n\tEmbedding\x12\x17.backend.PredictOptions\x1a\x18.backend.EmbeddingResult\"\x00\x12\x41\n\rGenerateImage\x12\x1d.backend.GenerateImageRequest\x1a\x0f.backend.Result\"\x00\x12M\n\x12\x41udioTranscription\x12\x1a.backend.TranscriptRequest\x1a\x19.backend.TranscriptResult\"\x00\x12-\n\x03TTS\x12\x13.backend.TTSRequest\x1a\x0f.backend.Result\"\x00\x42Z\n\x19io.skynet.localai.backendB\x0eLocalAIBackendP\x01Z+github.com/go-skynet/LocalAI/pkg/grpc/protob\x06proto3') + +_globals = globals() +_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals) +_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'backend_pb2', _globals) +if _descriptor._USE_C_DESCRIPTORS == False: + + DESCRIPTOR._options = None + DESCRIPTOR._serialized_options = b'\n\031io.skynet.localai.backendB\016LocalAIBackendP\001Z+github.com/go-skynet/LocalAI/pkg/grpc/proto' + _globals['_HEALTHMESSAGE']._serialized_start=26 + _globals['_HEALTHMESSAGE']._serialized_end=41 + _globals['_PREDICTOPTIONS']._serialized_start=44 + _globals['_PREDICTOPTIONS']._serialized_end=818 + _globals['_REPLY']._serialized_start=820 + _globals['_REPLY']._serialized_end=844 + _globals['_MODELOPTIONS']._serialized_start=847 + _globals['_MODELOPTIONS']._serialized_end=1329 + _globals['_RESULT']._serialized_start=1331 + _globals['_RESULT']._serialized_end=1373 + _globals['_EMBEDDINGRESULT']._serialized_start=1375 + _globals['_EMBEDDINGRESULT']._serialized_end=1412 + _globals['_TRANSCRIPTREQUEST']._serialized_start=1414 + _globals['_TRANSCRIPTREQUEST']._serialized_end=1481 + _globals['_TRANSCRIPTRESULT']._serialized_start=1483 + _globals['_TRANSCRIPTRESULT']._serialized_end=1561 + _globals['_TRANSCRIPTSEGMENT']._serialized_start=1563 + _globals['_TRANSCRIPTSEGMENT']._serialized_end=1652 + _globals['_GENERATEIMAGEREQUEST']._serialized_start=1655 + _globals['_GENERATEIMAGEREQUEST']._serialized_end=1813 + _globals['_TTSREQUEST']._serialized_start=1815 + _globals['_TTSREQUEST']._serialized_end=1869 + _globals['_BACKEND']._serialized_start=1872 + _globals['_BACKEND']._serialized_end=2363 +# @@protoc_insertion_point(module_scope) diff --git a/extra/grpc/bark/backend_pb2_grpc.py b/extra/grpc/bark/backend_pb2_grpc.py new file mode 100644 index 000000000..301c0729a --- /dev/null +++ b/extra/grpc/bark/backend_pb2_grpc.py @@ -0,0 +1,297 @@ +# Generated by the gRPC Python protocol compiler plugin. DO NOT EDIT! +"""Client and server classes corresponding to protobuf-defined services.""" +import grpc + +import backend_pb2 as backend__pb2 + + +class BackendStub(object): + """Missing associated documentation comment in .proto file.""" + + def __init__(self, channel): + """Constructor. + + Args: + channel: A grpc.Channel. + """ + self.Health = channel.unary_unary( + '/backend.Backend/Health', + request_serializer=backend__pb2.HealthMessage.SerializeToString, + response_deserializer=backend__pb2.Reply.FromString, + ) + self.Predict = channel.unary_unary( + '/backend.Backend/Predict', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.Reply.FromString, + ) + self.LoadModel = channel.unary_unary( + '/backend.Backend/LoadModel', + request_serializer=backend__pb2.ModelOptions.SerializeToString, + response_deserializer=backend__pb2.Result.FromString, + ) + self.PredictStream = channel.unary_stream( + '/backend.Backend/PredictStream', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.Reply.FromString, + ) + self.Embedding = channel.unary_unary( + '/backend.Backend/Embedding', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.EmbeddingResult.FromString, + ) + self.GenerateImage = channel.unary_unary( + '/backend.Backend/GenerateImage', + request_serializer=backend__pb2.GenerateImageRequest.SerializeToString, + response_deserializer=backend__pb2.Result.FromString, + ) + self.AudioTranscription = channel.unary_unary( + '/backend.Backend/AudioTranscription', + request_serializer=backend__pb2.TranscriptRequest.SerializeToString, + response_deserializer=backend__pb2.TranscriptResult.FromString, + ) + self.TTS = channel.unary_unary( + '/backend.Backend/TTS', + request_serializer=backend__pb2.TTSRequest.SerializeToString, + response_deserializer=backend__pb2.Result.FromString, + ) + + +class BackendServicer(object): + """Missing associated documentation comment in .proto file.""" + + def Health(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def Predict(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def LoadModel(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def PredictStream(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def Embedding(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def GenerateImage(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def AudioTranscription(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + def TTS(self, request, context): + """Missing associated documentation comment in .proto file.""" + context.set_code(grpc.StatusCode.UNIMPLEMENTED) + context.set_details('Method not implemented!') + raise NotImplementedError('Method not implemented!') + + +def add_BackendServicer_to_server(servicer, server): + rpc_method_handlers = { + 'Health': grpc.unary_unary_rpc_method_handler( + servicer.Health, + request_deserializer=backend__pb2.HealthMessage.FromString, + response_serializer=backend__pb2.Reply.SerializeToString, + ), + 'Predict': grpc.unary_unary_rpc_method_handler( + servicer.Predict, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.Reply.SerializeToString, + ), + 'LoadModel': grpc.unary_unary_rpc_method_handler( + servicer.LoadModel, + request_deserializer=backend__pb2.ModelOptions.FromString, + response_serializer=backend__pb2.Result.SerializeToString, + ), + 'PredictStream': grpc.unary_stream_rpc_method_handler( + servicer.PredictStream, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.Reply.SerializeToString, + ), + 'Embedding': grpc.unary_unary_rpc_method_handler( + servicer.Embedding, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.EmbeddingResult.SerializeToString, + ), + 'GenerateImage': grpc.unary_unary_rpc_method_handler( + servicer.GenerateImage, + request_deserializer=backend__pb2.GenerateImageRequest.FromString, + response_serializer=backend__pb2.Result.SerializeToString, + ), + 'AudioTranscription': grpc.unary_unary_rpc_method_handler( + servicer.AudioTranscription, + request_deserializer=backend__pb2.TranscriptRequest.FromString, + response_serializer=backend__pb2.TranscriptResult.SerializeToString, + ), + 'TTS': grpc.unary_unary_rpc_method_handler( + servicer.TTS, + request_deserializer=backend__pb2.TTSRequest.FromString, + response_serializer=backend__pb2.Result.SerializeToString, + ), + } + generic_handler = grpc.method_handlers_generic_handler( + 'backend.Backend', rpc_method_handlers) + server.add_generic_rpc_handlers((generic_handler,)) + + + # This class is part of an EXPERIMENTAL API. +class Backend(object): + """Missing associated documentation comment in .proto file.""" + + @staticmethod + def Health(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/Health', + backend__pb2.HealthMessage.SerializeToString, + backend__pb2.Reply.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def Predict(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/Predict', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.Reply.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def LoadModel(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/LoadModel', + backend__pb2.ModelOptions.SerializeToString, + backend__pb2.Result.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def PredictStream(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_stream(request, target, '/backend.Backend/PredictStream', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.Reply.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def Embedding(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/Embedding', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.EmbeddingResult.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def GenerateImage(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/GenerateImage', + backend__pb2.GenerateImageRequest.SerializeToString, + backend__pb2.Result.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def AudioTranscription(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/AudioTranscription', + backend__pb2.TranscriptRequest.SerializeToString, + backend__pb2.TranscriptResult.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def TTS(request, + target, + options=(), + channel_credentials=None, + call_credentials=None, + insecure=False, + compression=None, + wait_for_ready=None, + timeout=None, + metadata=None): + return grpc.experimental.unary_unary(request, target, '/backend.Backend/TTS', + backend__pb2.TTSRequest.SerializeToString, + backend__pb2.Result.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) diff --git a/extra/grpc/bark/ttsbark.py b/extra/grpc/bark/ttsbark.py new file mode 100644 index 000000000..63b01798e --- /dev/null +++ b/extra/grpc/bark/ttsbark.py @@ -0,0 +1,83 @@ +#!/usr/bin/env python3 +import grpc +from concurrent import futures +import time +import backend_pb2 +import backend_pb2_grpc +import argparse +import signal +import sys +import os +from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig +from pathlib import Path +from bark import SAMPLE_RATE, generate_audio, preload_models +from scipy.io.wavfile import write as write_wav + +_ONE_DAY_IN_SECONDS = 60 * 60 * 24 + +# Implement the BackendServicer class with the service methods +class BackendServicer(backend_pb2_grpc.BackendServicer): + def Health(self, request, context): + return backend_pb2.Reply(message=bytes("OK", 'utf-8')) + def LoadModel(self, request, context): + model_name = request.Model + try: + print("Preparing models, please wait", file=sys.stderr) + # download and load all models + preload_models() + except Exception as err: + return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}") + # Implement your logic here for the LoadModel service + # Replace this with your desired response + return backend_pb2.Result(message="Model loaded successfully", success=True) + + def TTS(self, request, context): + model = request.model + print(request, file=sys.stderr) + try: + audio_array = None + if model != "": + audio_array = generate_audio(request.text, history_prompt=model) + else: + audio_array = generate_audio(request.text) + print("saving to", request.dst, file=sys.stderr) + # save audio to disk + write_wav(request.dst, SAMPLE_RATE, audio_array) + print("saved to", request.dst, file=sys.stderr) + print("tts for", file=sys.stderr) + print(request, file=sys.stderr) + except Exception as err: + return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}") + return backend_pb2.Result(success=True) + +def serve(address): + server = grpc.server(futures.ThreadPoolExecutor(max_workers=10)) + backend_pb2_grpc.add_BackendServicer_to_server(BackendServicer(), server) + server.add_insecure_port(address) + server.start() + print("Server started. Listening on: " + address, file=sys.stderr) + + # Define the signal handler function + def signal_handler(sig, frame): + print("Received termination signal. Shutting down...") + server.stop(0) + sys.exit(0) + + # Set the signal handlers for SIGINT and SIGTERM + signal.signal(signal.SIGINT, signal_handler) + signal.signal(signal.SIGTERM, signal_handler) + + try: + while True: + time.sleep(_ONE_DAY_IN_SECONDS) + except KeyboardInterrupt: + server.stop(0) + +if __name__ == "__main__": + parser = argparse.ArgumentParser(description="Run the gRPC server.") + parser.add_argument( + "--addr", default="localhost:50051", help="The address to bind the server to." + ) + args = parser.parse_args() + + serve(args.addr) \ No newline at end of file diff --git a/extra/grpc/huggingface/backend_pb2.py b/extra/grpc/huggingface/backend_pb2.py index c062cee82..e8d2d30c8 100644 --- a/extra/grpc/huggingface/backend_pb2.py +++ b/extra/grpc/huggingface/backend_pb2.py @@ -13,7 +13,7 @@ _sym_db = _symbol_database.Default() -DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\rbackend.proto\x12\x07\x62\x61\x63kend\"\x0f\n\rHealthMessage\"\x86\x06\n\x0ePredictOptions\x12\x0e\n\x06Prompt\x18\x01 \x01(\t\x12\x0c\n\x04Seed\x18\x02 \x01(\x05\x12\x0f\n\x07Threads\x18\x03 \x01(\x05\x12\x0e\n\x06Tokens\x18\x04 \x01(\x05\x12\x0c\n\x04TopK\x18\x05 \x01(\x05\x12\x0e\n\x06Repeat\x18\x06 \x01(\x05\x12\r\n\x05\x42\x61tch\x18\x07 \x01(\x05\x12\r\n\x05NKeep\x18\x08 \x01(\x05\x12\x13\n\x0bTemperature\x18\t \x01(\x02\x12\x0f\n\x07Penalty\x18\n \x01(\x02\x12\r\n\x05\x46\x31\x36KV\x18\x0b \x01(\x08\x12\x11\n\tDebugMode\x18\x0c \x01(\x08\x12\x13\n\x0bStopPrompts\x18\r \x03(\t\x12\x11\n\tIgnoreEOS\x18\x0e \x01(\x08\x12\x19\n\x11TailFreeSamplingZ\x18\x0f \x01(\x02\x12\x10\n\x08TypicalP\x18\x10 \x01(\x02\x12\x18\n\x10\x46requencyPenalty\x18\x11 \x01(\x02\x12\x17\n\x0fPresencePenalty\x18\x12 \x01(\x02\x12\x10\n\x08Mirostat\x18\x13 \x01(\x05\x12\x13\n\x0bMirostatETA\x18\x14 \x01(\x02\x12\x13\n\x0bMirostatTAU\x18\x15 \x01(\x02\x12\x12\n\nPenalizeNL\x18\x16 \x01(\x08\x12\x11\n\tLogitBias\x18\x17 \x01(\t\x12\r\n\x05MLock\x18\x19 \x01(\x08\x12\x0c\n\x04MMap\x18\x1a \x01(\x08\x12\x16\n\x0ePromptCacheAll\x18\x1b \x01(\x08\x12\x15\n\rPromptCacheRO\x18\x1c \x01(\x08\x12\x0f\n\x07Grammar\x18\x1d \x01(\t\x12\x0f\n\x07MainGPU\x18\x1e \x01(\t\x12\x13\n\x0bTensorSplit\x18\x1f \x01(\t\x12\x0c\n\x04TopP\x18 \x01(\x02\x12\x17\n\x0fPromptCachePath\x18! \x01(\t\x12\r\n\x05\x44\x65\x62ug\x18\" \x01(\x08\x12\x17\n\x0f\x45mbeddingTokens\x18# \x03(\x05\x12\x12\n\nEmbeddings\x18$ \x01(\t\x12\x14\n\x0cRopeFreqBase\x18% \x01(\x02\x12\x15\n\rRopeFreqScale\x18& \x01(\x02\x12\x1b\n\x13NegativePromptScale\x18\' \x01(\x02\x12\x16\n\x0eNegativePrompt\x18( \x01(\t\"\x18\n\x05Reply\x12\x0f\n\x07message\x18\x01 \x01(\x0c\"\xfb\x02\n\x0cModelOptions\x12\r\n\x05Model\x18\x01 \x01(\t\x12\x13\n\x0b\x43ontextSize\x18\x02 \x01(\x05\x12\x0c\n\x04Seed\x18\x03 \x01(\x05\x12\x0e\n\x06NBatch\x18\x04 \x01(\x05\x12\x11\n\tF16Memory\x18\x05 \x01(\x08\x12\r\n\x05MLock\x18\x06 \x01(\x08\x12\x0c\n\x04MMap\x18\x07 \x01(\x08\x12\x11\n\tVocabOnly\x18\x08 \x01(\x08\x12\x0f\n\x07LowVRAM\x18\t \x01(\x08\x12\x12\n\nEmbeddings\x18\n \x01(\x08\x12\x0c\n\x04NUMA\x18\x0b \x01(\x08\x12\x12\n\nNGPULayers\x18\x0c \x01(\x05\x12\x0f\n\x07MainGPU\x18\r \x01(\t\x12\x13\n\x0bTensorSplit\x18\x0e \x01(\t\x12\x0f\n\x07Threads\x18\x0f \x01(\x05\x12\x19\n\x11LibrarySearchPath\x18\x10 \x01(\t\x12\x14\n\x0cRopeFreqBase\x18\x11 \x01(\x02\x12\x15\n\rRopeFreqScale\x18\x12 \x01(\x02\x12\x12\n\nRMSNormEps\x18\x13 \x01(\x02\x12\x0c\n\x04NGQA\x18\x14 \x01(\x05\"*\n\x06Result\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0f\n\x07success\x18\x02 \x01(\x08\"%\n\x0f\x45mbeddingResult\x12\x12\n\nembeddings\x18\x01 \x03(\x02\"C\n\x11TranscriptRequest\x12\x0b\n\x03\x64st\x18\x02 \x01(\t\x12\x10\n\x08language\x18\x03 \x01(\t\x12\x0f\n\x07threads\x18\x04 \x01(\r\"N\n\x10TranscriptResult\x12,\n\x08segments\x18\x01 \x03(\x0b\x32\x1a.backend.TranscriptSegment\x12\x0c\n\x04text\x18\x02 \x01(\t\"Y\n\x11TranscriptSegment\x12\n\n\x02id\x18\x01 \x01(\x05\x12\r\n\x05start\x18\x02 \x01(\x03\x12\x0b\n\x03\x65nd\x18\x03 \x01(\x03\x12\x0c\n\x04text\x18\x04 \x01(\t\x12\x0e\n\x06tokens\x18\x05 \x03(\x05\"\x9e\x01\n\x14GenerateImageRequest\x12\x0e\n\x06height\x18\x01 \x01(\x05\x12\r\n\x05width\x18\x02 \x01(\x05\x12\x0c\n\x04mode\x18\x03 \x01(\x05\x12\x0c\n\x04step\x18\x04 \x01(\x05\x12\x0c\n\x04seed\x18\x05 \x01(\x05\x12\x17\n\x0fpositive_prompt\x18\x06 \x01(\t\x12\x17\n\x0fnegative_prompt\x18\x07 \x01(\t\x12\x0b\n\x03\x64st\x18\x08 \x01(\t\"6\n\nTTSRequest\x12\x0c\n\x04text\x18\x01 \x01(\t\x12\r\n\x05model\x18\x02 \x01(\t\x12\x0b\n\x03\x64st\x18\x03 \x01(\t2\xeb\x03\n\x07\x42\x61\x63kend\x12\x32\n\x06Health\x12\x16.backend.HealthMessage\x1a\x0e.backend.Reply\"\x00\x12\x34\n\x07Predict\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x12\x35\n\tLoadModel\x12\x15.backend.ModelOptions\x1a\x0f.backend.Result\"\x00\x12<\n\rPredictStream\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x30\x01\x12@\n\tEmbedding\x12\x17.backend.PredictOptions\x1a\x18.backend.EmbeddingResult\"\x00\x12\x41\n\rGenerateImage\x12\x1d.backend.GenerateImageRequest\x1a\x0f.backend.Result\"\x00\x12M\n\x12\x41udioTranscription\x12\x1a.backend.TranscriptRequest\x1a\x19.backend.TranscriptResult\"\x00\x12-\n\x03TTS\x12\x13.backend.TTSRequest\x1a\x0f.backend.Result\"\x00\x42Z\n\x19io.skynet.localai.backendB\x0eLocalAIBackendP\x01Z+github.com/go-skynet/LocalAI/pkg/grpc/protob\x06proto3') +DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\rbackend.proto\x12\x07\x62\x61\x63kend\"\x0f\n\rHealthMessage\"\x86\x06\n\x0ePredictOptions\x12\x0e\n\x06Prompt\x18\x01 \x01(\t\x12\x0c\n\x04Seed\x18\x02 \x01(\x05\x12\x0f\n\x07Threads\x18\x03 \x01(\x05\x12\x0e\n\x06Tokens\x18\x04 \x01(\x05\x12\x0c\n\x04TopK\x18\x05 \x01(\x05\x12\x0e\n\x06Repeat\x18\x06 \x01(\x05\x12\r\n\x05\x42\x61tch\x18\x07 \x01(\x05\x12\r\n\x05NKeep\x18\x08 \x01(\x05\x12\x13\n\x0bTemperature\x18\t \x01(\x02\x12\x0f\n\x07Penalty\x18\n \x01(\x02\x12\r\n\x05\x46\x31\x36KV\x18\x0b \x01(\x08\x12\x11\n\tDebugMode\x18\x0c \x01(\x08\x12\x13\n\x0bStopPrompts\x18\r \x03(\t\x12\x11\n\tIgnoreEOS\x18\x0e \x01(\x08\x12\x19\n\x11TailFreeSamplingZ\x18\x0f \x01(\x02\x12\x10\n\x08TypicalP\x18\x10 \x01(\x02\x12\x18\n\x10\x46requencyPenalty\x18\x11 \x01(\x02\x12\x17\n\x0fPresencePenalty\x18\x12 \x01(\x02\x12\x10\n\x08Mirostat\x18\x13 \x01(\x05\x12\x13\n\x0bMirostatETA\x18\x14 \x01(\x02\x12\x13\n\x0bMirostatTAU\x18\x15 \x01(\x02\x12\x12\n\nPenalizeNL\x18\x16 \x01(\x08\x12\x11\n\tLogitBias\x18\x17 \x01(\t\x12\r\n\x05MLock\x18\x19 \x01(\x08\x12\x0c\n\x04MMap\x18\x1a \x01(\x08\x12\x16\n\x0ePromptCacheAll\x18\x1b \x01(\x08\x12\x15\n\rPromptCacheRO\x18\x1c \x01(\x08\x12\x0f\n\x07Grammar\x18\x1d \x01(\t\x12\x0f\n\x07MainGPU\x18\x1e \x01(\t\x12\x13\n\x0bTensorSplit\x18\x1f \x01(\t\x12\x0c\n\x04TopP\x18 \x01(\x02\x12\x17\n\x0fPromptCachePath\x18! \x01(\t\x12\r\n\x05\x44\x65\x62ug\x18\" \x01(\x08\x12\x17\n\x0f\x45mbeddingTokens\x18# \x03(\x05\x12\x12\n\nEmbeddings\x18$ \x01(\t\x12\x14\n\x0cRopeFreqBase\x18% \x01(\x02\x12\x15\n\rRopeFreqScale\x18& \x01(\x02\x12\x1b\n\x13NegativePromptScale\x18\' \x01(\x02\x12\x16\n\x0eNegativePrompt\x18( \x01(\t\"\x18\n\x05Reply\x12\x0f\n\x07message\x18\x01 \x01(\x0c\"\xe2\x03\n\x0cModelOptions\x12\r\n\x05Model\x18\x01 \x01(\t\x12\x13\n\x0b\x43ontextSize\x18\x02 \x01(\x05\x12\x0c\n\x04Seed\x18\x03 \x01(\x05\x12\x0e\n\x06NBatch\x18\x04 \x01(\x05\x12\x11\n\tF16Memory\x18\x05 \x01(\x08\x12\r\n\x05MLock\x18\x06 \x01(\x08\x12\x0c\n\x04MMap\x18\x07 \x01(\x08\x12\x11\n\tVocabOnly\x18\x08 \x01(\x08\x12\x0f\n\x07LowVRAM\x18\t \x01(\x08\x12\x12\n\nEmbeddings\x18\n \x01(\x08\x12\x0c\n\x04NUMA\x18\x0b \x01(\x08\x12\x12\n\nNGPULayers\x18\x0c \x01(\x05\x12\x0f\n\x07MainGPU\x18\r \x01(\t\x12\x13\n\x0bTensorSplit\x18\x0e \x01(\t\x12\x0f\n\x07Threads\x18\x0f \x01(\x05\x12\x19\n\x11LibrarySearchPath\x18\x10 \x01(\t\x12\x14\n\x0cRopeFreqBase\x18\x11 \x01(\x02\x12\x15\n\rRopeFreqScale\x18\x12 \x01(\x02\x12\x12\n\nRMSNormEps\x18\x13 \x01(\x02\x12\x0c\n\x04NGQA\x18\x14 \x01(\x05\x12\x11\n\tModelFile\x18\x15 \x01(\t\x12\x0e\n\x06\x44\x65vice\x18\x16 \x01(\t\x12\x11\n\tUseTriton\x18\x17 \x01(\x08\x12\x15\n\rModelBaseName\x18\x18 \x01(\t\x12\x18\n\x10UseFastTokenizer\x18\x19 \x01(\x08\"*\n\x06Result\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0f\n\x07success\x18\x02 \x01(\x08\"%\n\x0f\x45mbeddingResult\x12\x12\n\nembeddings\x18\x01 \x03(\x02\"C\n\x11TranscriptRequest\x12\x0b\n\x03\x64st\x18\x02 \x01(\t\x12\x10\n\x08language\x18\x03 \x01(\t\x12\x0f\n\x07threads\x18\x04 \x01(\r\"N\n\x10TranscriptResult\x12,\n\x08segments\x18\x01 \x03(\x0b\x32\x1a.backend.TranscriptSegment\x12\x0c\n\x04text\x18\x02 \x01(\t\"Y\n\x11TranscriptSegment\x12\n\n\x02id\x18\x01 \x01(\x05\x12\r\n\x05start\x18\x02 \x01(\x03\x12\x0b\n\x03\x65nd\x18\x03 \x01(\x03\x12\x0c\n\x04text\x18\x04 \x01(\t\x12\x0e\n\x06tokens\x18\x05 \x03(\x05\"\x9e\x01\n\x14GenerateImageRequest\x12\x0e\n\x06height\x18\x01 \x01(\x05\x12\r\n\x05width\x18\x02 \x01(\x05\x12\x0c\n\x04mode\x18\x03 \x01(\x05\x12\x0c\n\x04step\x18\x04 \x01(\x05\x12\x0c\n\x04seed\x18\x05 \x01(\x05\x12\x17\n\x0fpositive_prompt\x18\x06 \x01(\t\x12\x17\n\x0fnegative_prompt\x18\x07 \x01(\t\x12\x0b\n\x03\x64st\x18\x08 \x01(\t\"6\n\nTTSRequest\x12\x0c\n\x04text\x18\x01 \x01(\t\x12\r\n\x05model\x18\x02 \x01(\t\x12\x0b\n\x03\x64st\x18\x03 \x01(\t2\xeb\x03\n\x07\x42\x61\x63kend\x12\x32\n\x06Health\x12\x16.backend.HealthMessage\x1a\x0e.backend.Reply\"\x00\x12\x34\n\x07Predict\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x12\x35\n\tLoadModel\x12\x15.backend.ModelOptions\x1a\x0f.backend.Result\"\x00\x12<\n\rPredictStream\x12\x17.backend.PredictOptions\x1a\x0e.backend.Reply\"\x00\x30\x01\x12@\n\tEmbedding\x12\x17.backend.PredictOptions\x1a\x18.backend.EmbeddingResult\"\x00\x12\x41\n\rGenerateImage\x12\x1d.backend.GenerateImageRequest\x1a\x0f.backend.Result\"\x00\x12M\n\x12\x41udioTranscription\x12\x1a.backend.TranscriptRequest\x1a\x19.backend.TranscriptResult\"\x00\x12-\n\x03TTS\x12\x13.backend.TTSRequest\x1a\x0f.backend.Result\"\x00\x42Z\n\x19io.skynet.localai.backendB\x0eLocalAIBackendP\x01Z+github.com/go-skynet/LocalAI/pkg/grpc/protob\x06proto3') _globals = globals() _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals) @@ -29,21 +29,21 @@ if _descriptor._USE_C_DESCRIPTORS == False: _globals['_REPLY']._serialized_start=820 _globals['_REPLY']._serialized_end=844 _globals['_MODELOPTIONS']._serialized_start=847 - _globals['_MODELOPTIONS']._serialized_end=1226 - _globals['_RESULT']._serialized_start=1228 - _globals['_RESULT']._serialized_end=1270 - _globals['_EMBEDDINGRESULT']._serialized_start=1272 - _globals['_EMBEDDINGRESULT']._serialized_end=1309 - _globals['_TRANSCRIPTREQUEST']._serialized_start=1311 - _globals['_TRANSCRIPTREQUEST']._serialized_end=1378 - _globals['_TRANSCRIPTRESULT']._serialized_start=1380 - _globals['_TRANSCRIPTRESULT']._serialized_end=1458 - _globals['_TRANSCRIPTSEGMENT']._serialized_start=1460 - _globals['_TRANSCRIPTSEGMENT']._serialized_end=1549 - _globals['_GENERATEIMAGEREQUEST']._serialized_start=1552 - _globals['_GENERATEIMAGEREQUEST']._serialized_end=1710 - _globals['_TTSREQUEST']._serialized_start=1712 - _globals['_TTSREQUEST']._serialized_end=1766 - _globals['_BACKEND']._serialized_start=1769 - _globals['_BACKEND']._serialized_end=2260 + _globals['_MODELOPTIONS']._serialized_end=1329 + _globals['_RESULT']._serialized_start=1331 + _globals['_RESULT']._serialized_end=1373 + _globals['_EMBEDDINGRESULT']._serialized_start=1375 + _globals['_EMBEDDINGRESULT']._serialized_end=1412 + _globals['_TRANSCRIPTREQUEST']._serialized_start=1414 + _globals['_TRANSCRIPTREQUEST']._serialized_end=1481 + _globals['_TRANSCRIPTRESULT']._serialized_start=1483 + _globals['_TRANSCRIPTRESULT']._serialized_end=1561 + _globals['_TRANSCRIPTSEGMENT']._serialized_start=1563 + _globals['_TRANSCRIPTSEGMENT']._serialized_end=1652 + _globals['_GENERATEIMAGEREQUEST']._serialized_start=1655 + _globals['_GENERATEIMAGEREQUEST']._serialized_end=1813 + _globals['_TTSREQUEST']._serialized_start=1815 + _globals['_TTSREQUEST']._serialized_end=1869 + _globals['_BACKEND']._serialized_start=1872 + _globals['_BACKEND']._serialized_end=2363 # @@protoc_insertion_point(module_scope) diff --git a/extra/grpc/huggingface/huggingface.py b/extra/grpc/huggingface/huggingface.py index 047769309..7589dfd67 100755 --- a/extra/grpc/huggingface/huggingface.py +++ b/extra/grpc/huggingface/huggingface.py @@ -18,7 +18,6 @@ class BackendServicer(backend_pb2_grpc.BackendServicer): return backend_pb2.Reply(message=bytes("OK", 'utf-8')) def LoadModel(self, request, context): model_name = request.Model - model_name = os.path.basename(model_name) try: self.model = SentenceTransformer(model_name) except Exception as err: diff --git a/extra/requirements.txt b/extra/requirements.txt index 9744afb16..183b6badc 100644 --- a/extra/requirements.txt +++ b/extra/requirements.txt @@ -1,4 +1,6 @@ sentence_transformers grpcio google -protobuf \ No newline at end of file +protobuf +torch +git+https://github.com/suno-ai/bark.git \ No newline at end of file diff --git a/pkg/gallery/gallery.go b/pkg/gallery/gallery.go index f9a140023..fbc59d627 100644 --- a/pkg/gallery/gallery.go +++ b/pkg/gallery/gallery.go @@ -85,7 +85,7 @@ func InstallModelFromGalleryByName(galleries []Gallery, name string, basePath st name = strings.ReplaceAll(name, string(os.PathSeparator), "__") var model *GalleryModel for _, m := range models { - if name == m.Name || name == strings.ToLower(m.Name) { + if name == m.Name || m.Name == strings.ToLower(name) { model = m } } diff --git a/pkg/grpc/image/stablediffusion.go b/pkg/grpc/image/stablediffusion.go index ce0275e90..600f20071 100644 --- a/pkg/grpc/image/stablediffusion.go +++ b/pkg/grpc/image/stablediffusion.go @@ -16,7 +16,7 @@ type StableDiffusion struct { func (sd *StableDiffusion) Load(opts *pb.ModelOptions) error { var err error // Note: the Model here is a path to a directory containing the model files - sd.stablediffusion, err = stablediffusion.New(opts.Model) + sd.stablediffusion, err = stablediffusion.New(opts.ModelFile) return err } diff --git a/pkg/grpc/llm/bert/bert.go b/pkg/grpc/llm/bert/bert.go index 7692797e6..b7df5d768 100644 --- a/pkg/grpc/llm/bert/bert.go +++ b/pkg/grpc/llm/bert/bert.go @@ -15,7 +15,7 @@ type Embeddings struct { } func (llm *Embeddings) Load(opts *pb.ModelOptions) error { - model, err := bert.New(opts.Model) + model, err := bert.New(opts.ModelFile) llm.bert = model return err } diff --git a/pkg/grpc/llm/bloomz/bloomz.go b/pkg/grpc/llm/bloomz/bloomz.go index daa226407..fecaa5b79 100644 --- a/pkg/grpc/llm/bloomz/bloomz.go +++ b/pkg/grpc/llm/bloomz/bloomz.go @@ -18,7 +18,7 @@ type LLM struct { } func (llm *LLM) Load(opts *pb.ModelOptions) error { - model, err := bloomz.New(opts.Model) + model, err := bloomz.New(opts.ModelFile) llm.bloomz = model return err } diff --git a/pkg/grpc/llm/falcon/falcon.go b/pkg/grpc/llm/falcon/falcon.go index 3c0f84edc..d287ef6dc 100644 --- a/pkg/grpc/llm/falcon/falcon.go +++ b/pkg/grpc/llm/falcon/falcon.go @@ -40,7 +40,7 @@ func (llm *LLM) Load(opts *pb.ModelOptions) error { ggllmOpts = append(ggllmOpts, ggllm.SetNBatch(512)) } - model, err := ggllm.New(opts.Model, ggllmOpts...) + model, err := ggllm.New(opts.ModelFile, ggllmOpts...) llm.falcon = model return err } diff --git a/pkg/grpc/llm/gpt4all/gpt4all.go b/pkg/grpc/llm/gpt4all/gpt4all.go index e17afc1ef..f94e33094 100644 --- a/pkg/grpc/llm/gpt4all/gpt4all.go +++ b/pkg/grpc/llm/gpt4all/gpt4all.go @@ -17,7 +17,7 @@ type LLM struct { } func (llm *LLM) Load(opts *pb.ModelOptions) error { - model, err := gpt4all.New(opts.Model, + model, err := gpt4all.New(opts.ModelFile, gpt4all.SetThreads(int(opts.Threads)), gpt4all.SetLibrarySearchPath(opts.LibrarySearchPath)) llm.gpt4all = model diff --git a/pkg/grpc/llm/llama/llama.go b/pkg/grpc/llm/llama/llama.go index 79e550488..da46c9f57 100644 --- a/pkg/grpc/llm/llama/llama.go +++ b/pkg/grpc/llm/llama/llama.go @@ -71,7 +71,7 @@ func (llm *LLM) Load(opts *pb.ModelOptions) error { llamaOpts = append(llamaOpts, llama.EnabelLowVRAM) } - model, err := llama.New(opts.Model, llamaOpts...) + model, err := llama.New(opts.ModelFile, llamaOpts...) llm.llama = model return err } diff --git a/pkg/grpc/llm/rwkv/rwkv.go b/pkg/grpc/llm/rwkv/rwkv.go index f54c14bd0..cfd60f0c0 100644 --- a/pkg/grpc/llm/rwkv/rwkv.go +++ b/pkg/grpc/llm/rwkv/rwkv.go @@ -20,9 +20,9 @@ type LLM struct { } func (llm *LLM) Load(opts *pb.ModelOptions) error { - modelPath := filepath.Dir(opts.Model) - modelFile := filepath.Base(opts.Model) - model := rwkv.LoadFiles(opts.Model, filepath.Join(modelPath, modelFile+tokenizerSuffix), uint32(opts.GetThreads())) + modelPath := filepath.Dir(opts.ModelFile) + modelFile := filepath.Base(opts.ModelFile) + model := rwkv.LoadFiles(opts.ModelFile, filepath.Join(modelPath, modelFile+tokenizerSuffix), uint32(opts.GetThreads())) if model == nil { return fmt.Errorf("could not load model") diff --git a/pkg/grpc/llm/transformers/dolly.go b/pkg/grpc/llm/transformers/dolly.go index d5f30938f..6c9e0a5df 100644 --- a/pkg/grpc/llm/transformers/dolly.go +++ b/pkg/grpc/llm/transformers/dolly.go @@ -18,7 +18,7 @@ type Dolly struct { } func (llm *Dolly) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewDolly(opts.Model) + model, err := transformers.NewDolly(opts.ModelFile) llm.dolly = model return err } diff --git a/pkg/grpc/llm/transformers/falcon.go b/pkg/grpc/llm/transformers/falcon.go index 982e43e03..54e9b3200 100644 --- a/pkg/grpc/llm/transformers/falcon.go +++ b/pkg/grpc/llm/transformers/falcon.go @@ -18,7 +18,7 @@ type Falcon struct { } func (llm *Falcon) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewFalcon(opts.Model) + model, err := transformers.NewFalcon(opts.ModelFile) llm.falcon = model return err } diff --git a/pkg/grpc/llm/transformers/gpt2.go b/pkg/grpc/llm/transformers/gpt2.go index 85a411254..66517d7c7 100644 --- a/pkg/grpc/llm/transformers/gpt2.go +++ b/pkg/grpc/llm/transformers/gpt2.go @@ -18,7 +18,7 @@ type GPT2 struct { } func (llm *GPT2) Load(opts *pb.ModelOptions) error { - model, err := transformers.New(opts.Model) + model, err := transformers.New(opts.ModelFile) llm.gpt2 = model return err } diff --git a/pkg/grpc/llm/transformers/gptj.go b/pkg/grpc/llm/transformers/gptj.go index e2bc3bf10..6f692188e 100644 --- a/pkg/grpc/llm/transformers/gptj.go +++ b/pkg/grpc/llm/transformers/gptj.go @@ -18,7 +18,7 @@ type GPTJ struct { } func (llm *GPTJ) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewGPTJ(opts.Model) + model, err := transformers.NewGPTJ(opts.ModelFile) llm.gptj = model return err } diff --git a/pkg/grpc/llm/transformers/gptneox.go b/pkg/grpc/llm/transformers/gptneox.go index ca6db941e..bcff0abec 100644 --- a/pkg/grpc/llm/transformers/gptneox.go +++ b/pkg/grpc/llm/transformers/gptneox.go @@ -18,7 +18,7 @@ type GPTNeoX struct { } func (llm *GPTNeoX) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewGPTNeoX(opts.Model) + model, err := transformers.NewGPTNeoX(opts.ModelFile) llm.gptneox = model return err } diff --git a/pkg/grpc/llm/transformers/mpt.go b/pkg/grpc/llm/transformers/mpt.go index d2b9ff1f4..8adda3614 100644 --- a/pkg/grpc/llm/transformers/mpt.go +++ b/pkg/grpc/llm/transformers/mpt.go @@ -18,7 +18,7 @@ type MPT struct { } func (llm *MPT) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewMPT(opts.Model) + model, err := transformers.NewMPT(opts.ModelFile) llm.mpt = model return err } diff --git a/pkg/grpc/llm/transformers/replit.go b/pkg/grpc/llm/transformers/replit.go index 4b26ffd84..bfc3a90ae 100644 --- a/pkg/grpc/llm/transformers/replit.go +++ b/pkg/grpc/llm/transformers/replit.go @@ -18,7 +18,7 @@ type Replit struct { } func (llm *Replit) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewReplit(opts.Model) + model, err := transformers.NewReplit(opts.ModelFile) llm.replit = model return err } diff --git a/pkg/grpc/llm/transformers/starcoder.go b/pkg/grpc/llm/transformers/starcoder.go index 7631274e3..06948cb3e 100644 --- a/pkg/grpc/llm/transformers/starcoder.go +++ b/pkg/grpc/llm/transformers/starcoder.go @@ -18,7 +18,7 @@ type Starcoder struct { } func (llm *Starcoder) Load(opts *pb.ModelOptions) error { - model, err := transformers.NewStarcoder(opts.Model) + model, err := transformers.NewStarcoder(opts.ModelFile) llm.starcoder = model return err } diff --git a/pkg/grpc/proto/backend.pb.go b/pkg/grpc/proto/backend.pb.go index 3a58b3491..33d09b461 100644 --- a/pkg/grpc/proto/backend.pb.go +++ b/pkg/grpc/proto/backend.pb.go @@ -483,6 +483,12 @@ type ModelOptions struct { RopeFreqScale float32 `protobuf:"fixed32,18,opt,name=RopeFreqScale,proto3" json:"RopeFreqScale,omitempty"` RMSNormEps float32 `protobuf:"fixed32,19,opt,name=RMSNormEps,proto3" json:"RMSNormEps,omitempty"` NGQA int32 `protobuf:"varint,20,opt,name=NGQA,proto3" json:"NGQA,omitempty"` + ModelFile string `protobuf:"bytes,21,opt,name=ModelFile,proto3" json:"ModelFile,omitempty"` + // AutoGPTQ + Device string `protobuf:"bytes,22,opt,name=Device,proto3" json:"Device,omitempty"` + UseTriton bool `protobuf:"varint,23,opt,name=UseTriton,proto3" json:"UseTriton,omitempty"` + ModelBaseName string `protobuf:"bytes,24,opt,name=ModelBaseName,proto3" json:"ModelBaseName,omitempty"` + UseFastTokenizer bool `protobuf:"varint,25,opt,name=UseFastTokenizer,proto3" json:"UseFastTokenizer,omitempty"` } func (x *ModelOptions) Reset() { @@ -657,6 +663,41 @@ func (x *ModelOptions) GetNGQA() int32 { return 0 } +func (x *ModelOptions) GetModelFile() string { + if x != nil { + return x.ModelFile + } + return "" +} + +func (x *ModelOptions) GetDevice() string { + if x != nil { + return x.Device + } + return "" +} + +func (x *ModelOptions) GetUseTriton() bool { + if x != nil { + return x.UseTriton + } + return false +} + +func (x *ModelOptions) GetModelBaseName() string { + if x != nil { + return x.ModelBaseName + } + return "" +} + +func (x *ModelOptions) GetUseFastTokenizer() bool { + if x != nil { + return x.UseFastTokenizer + } + return false +} + type Result struct { state protoimpl.MessageState sizeCache protoimpl.SizeCache @@ -1207,7 +1248,7 @@ var file_pkg_grpc_proto_backend_proto_rawDesc = []byte{ 0x0e, 0x4e, 0x65, 0x67, 0x61, 0x74, 0x69, 0x76, 0x65, 0x50, 0x72, 0x6f, 0x6d, 0x70, 0x74, 0x22, 0x21, 0x0a, 0x05, 0x52, 0x65, 0x70, 0x6c, 0x79, 0x12, 0x18, 0x0a, 0x07, 0x6d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x18, 0x01, 0x20, 0x01, 0x28, 0x0c, 0x52, 0x07, 0x6d, 0x65, 0x73, 0x73, 0x61, - 0x67, 0x65, 0x22, 0xc8, 0x04, 0x0a, 0x0c, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x4f, 0x70, 0x74, 0x69, + 0x67, 0x65, 0x22, 0xee, 0x05, 0x0a, 0x0c, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x4f, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x12, 0x14, 0x0a, 0x05, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x18, 0x01, 0x20, 0x01, 0x28, 0x09, 0x52, 0x05, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x12, 0x20, 0x0a, 0x0b, 0x43, 0x6f, 0x6e, 0x74, 0x65, 0x78, 0x74, 0x53, 0x69, 0x7a, 0x65, 0x18, 0x02, 0x20, 0x01, 0x28, 0x05, 0x52, 0x0b, @@ -1243,90 +1284,100 @@ var file_pkg_grpc_proto_backend_proto_rawDesc = []byte{ 0x46, 0x72, 0x65, 0x71, 0x53, 0x63, 0x61, 0x6c, 0x65, 0x12, 0x1e, 0x0a, 0x0a, 0x52, 0x4d, 0x53, 0x4e, 0x6f, 0x72, 0x6d, 0x45, 0x70, 0x73, 0x18, 0x13, 0x20, 0x01, 0x28, 0x02, 0x52, 0x0a, 0x52, 0x4d, 0x53, 0x4e, 0x6f, 0x72, 0x6d, 0x45, 0x70, 0x73, 0x12, 0x12, 0x0a, 0x04, 0x4e, 0x47, 0x51, - 0x41, 0x18, 0x14, 0x20, 0x01, 0x28, 0x05, 0x52, 0x04, 0x4e, 0x47, 0x51, 0x41, 0x22, 0x3c, 0x0a, - 0x06, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x12, 0x18, 0x0a, 0x07, 0x6d, 0x65, 0x73, 0x73, 0x61, - 0x67, 0x65, 0x18, 0x01, 0x20, 0x01, 0x28, 0x09, 0x52, 0x07, 0x6d, 0x65, 0x73, 0x73, 0x61, 0x67, - 0x65, 0x12, 0x18, 0x0a, 0x07, 0x73, 0x75, 0x63, 0x63, 0x65, 0x73, 0x73, 0x18, 0x02, 0x20, 0x01, - 0x28, 0x08, 0x52, 0x07, 0x73, 0x75, 0x63, 0x63, 0x65, 0x73, 0x73, 0x22, 0x31, 0x0a, 0x0f, 0x45, - 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x12, 0x1e, - 0x0a, 0x0a, 0x65, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, 0x73, 0x18, 0x01, 0x20, 0x03, - 0x28, 0x02, 0x52, 0x0a, 0x65, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, 0x73, 0x22, 0x5b, - 0x0a, 0x11, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x52, 0x65, 0x71, 0x75, - 0x65, 0x73, 0x74, 0x12, 0x10, 0x0a, 0x03, 0x64, 0x73, 0x74, 0x18, 0x02, 0x20, 0x01, 0x28, 0x09, - 0x52, 0x03, 0x64, 0x73, 0x74, 0x12, 0x1a, 0x0a, 0x08, 0x6c, 0x61, 0x6e, 0x67, 0x75, 0x61, 0x67, - 0x65, 0x18, 0x03, 0x20, 0x01, 0x28, 0x09, 0x52, 0x08, 0x6c, 0x61, 0x6e, 0x67, 0x75, 0x61, 0x67, - 0x65, 0x12, 0x18, 0x0a, 0x07, 0x74, 0x68, 0x72, 0x65, 0x61, 0x64, 0x73, 0x18, 0x04, 0x20, 0x01, - 0x28, 0x0d, 0x52, 0x07, 0x74, 0x68, 0x72, 0x65, 0x61, 0x64, 0x73, 0x22, 0x5e, 0x0a, 0x10, 0x54, - 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x12, - 0x36, 0x0a, 0x08, 0x73, 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, 0x73, 0x18, 0x01, 0x20, 0x03, 0x28, - 0x0b, 0x32, 0x1a, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x54, 0x72, 0x61, 0x6e, - 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x53, 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, 0x52, 0x08, 0x73, - 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, 0x73, 0x12, 0x12, 0x0a, 0x04, 0x74, 0x65, 0x78, 0x74, 0x18, - 0x02, 0x20, 0x01, 0x28, 0x09, 0x52, 0x04, 0x74, 0x65, 0x78, 0x74, 0x22, 0x77, 0x0a, 0x11, 0x54, - 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x53, 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, - 0x12, 0x0e, 0x0a, 0x02, 0x69, 0x64, 0x18, 0x01, 0x20, 0x01, 0x28, 0x05, 0x52, 0x02, 0x69, 0x64, - 0x12, 0x14, 0x0a, 0x05, 0x73, 0x74, 0x61, 0x72, 0x74, 0x18, 0x02, 0x20, 0x01, 0x28, 0x03, 0x52, - 0x05, 0x73, 0x74, 0x61, 0x72, 0x74, 0x12, 0x10, 0x0a, 0x03, 0x65, 0x6e, 0x64, 0x18, 0x03, 0x20, - 0x01, 0x28, 0x03, 0x52, 0x03, 0x65, 0x6e, 0x64, 0x12, 0x12, 0x0a, 0x04, 0x74, 0x65, 0x78, 0x74, - 0x18, 0x04, 0x20, 0x01, 0x28, 0x09, 0x52, 0x04, 0x74, 0x65, 0x78, 0x74, 0x12, 0x16, 0x0a, 0x06, - 0x74, 0x6f, 0x6b, 0x65, 0x6e, 0x73, 0x18, 0x05, 0x20, 0x03, 0x28, 0x05, 0x52, 0x06, 0x74, 0x6f, - 0x6b, 0x65, 0x6e, 0x73, 0x22, 0xe4, 0x01, 0x0a, 0x14, 0x47, 0x65, 0x6e, 0x65, 0x72, 0x61, 0x74, - 0x65, 0x49, 0x6d, 0x61, 0x67, 0x65, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x12, 0x16, 0x0a, - 0x06, 0x68, 0x65, 0x69, 0x67, 0x68, 0x74, 0x18, 0x01, 0x20, 0x01, 0x28, 0x05, 0x52, 0x06, 0x68, - 0x65, 0x69, 0x67, 0x68, 0x74, 0x12, 0x14, 0x0a, 0x05, 0x77, 0x69, 0x64, 0x74, 0x68, 0x18, 0x02, - 0x20, 0x01, 0x28, 0x05, 0x52, 0x05, 0x77, 0x69, 0x64, 0x74, 0x68, 0x12, 0x12, 0x0a, 0x04, 0x6d, - 0x6f, 0x64, 0x65, 0x18, 0x03, 0x20, 0x01, 0x28, 0x05, 0x52, 0x04, 0x6d, 0x6f, 0x64, 0x65, 0x12, - 0x12, 0x0a, 0x04, 0x73, 0x74, 0x65, 0x70, 0x18, 0x04, 0x20, 0x01, 0x28, 0x05, 0x52, 0x04, 0x73, - 0x74, 0x65, 0x70, 0x12, 0x12, 0x0a, 0x04, 0x73, 0x65, 0x65, 0x64, 0x18, 0x05, 0x20, 0x01, 0x28, - 0x05, 0x52, 0x04, 0x73, 0x65, 0x65, 0x64, 0x12, 0x27, 0x0a, 0x0f, 0x70, 0x6f, 0x73, 0x69, 0x74, - 0x69, 0x76, 0x65, 0x5f, 0x70, 0x72, 0x6f, 0x6d, 0x70, 0x74, 0x18, 0x06, 0x20, 0x01, 0x28, 0x09, - 0x52, 0x0e, 0x70, 0x6f, 0x73, 0x69, 0x74, 0x69, 0x76, 0x65, 0x50, 0x72, 0x6f, 0x6d, 0x70, 0x74, - 0x12, 0x27, 0x0a, 0x0f, 0x6e, 0x65, 0x67, 0x61, 0x74, 0x69, 0x76, 0x65, 0x5f, 0x70, 0x72, 0x6f, - 0x6d, 0x70, 0x74, 0x18, 0x07, 0x20, 0x01, 0x28, 0x09, 0x52, 0x0e, 0x6e, 0x65, 0x67, 0x61, 0x74, - 0x69, 0x76, 0x65, 0x50, 0x72, 0x6f, 0x6d, 0x70, 0x74, 0x12, 0x10, 0x0a, 0x03, 0x64, 0x73, 0x74, - 0x18, 0x08, 0x20, 0x01, 0x28, 0x09, 0x52, 0x03, 0x64, 0x73, 0x74, 0x22, 0x48, 0x0a, 0x0a, 0x54, - 0x54, 0x53, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x12, 0x12, 0x0a, 0x04, 0x74, 0x65, 0x78, - 0x74, 0x18, 0x01, 0x20, 0x01, 0x28, 0x09, 0x52, 0x04, 0x74, 0x65, 0x78, 0x74, 0x12, 0x14, 0x0a, - 0x05, 0x6d, 0x6f, 0x64, 0x65, 0x6c, 0x18, 0x02, 0x20, 0x01, 0x28, 0x09, 0x52, 0x05, 0x6d, 0x6f, - 0x64, 0x65, 0x6c, 0x12, 0x10, 0x0a, 0x03, 0x64, 0x73, 0x74, 0x18, 0x03, 0x20, 0x01, 0x28, 0x09, - 0x52, 0x03, 0x64, 0x73, 0x74, 0x32, 0xeb, 0x03, 0x0a, 0x07, 0x42, 0x61, 0x63, 0x6b, 0x65, 0x6e, - 0x64, 0x12, 0x32, 0x0a, 0x06, 0x48, 0x65, 0x61, 0x6c, 0x74, 0x68, 0x12, 0x16, 0x2e, 0x62, 0x61, - 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x48, 0x65, 0x61, 0x6c, 0x74, 0x68, 0x4d, 0x65, 0x73, 0x73, - 0x61, 0x67, 0x65, 0x1a, 0x0e, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, - 0x70, 0x6c, 0x79, 0x22, 0x00, 0x12, 0x34, 0x0a, 0x07, 0x50, 0x72, 0x65, 0x64, 0x69, 0x63, 0x74, - 0x12, 0x17, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x50, 0x72, 0x65, 0x64, 0x69, - 0x63, 0x74, 0x4f, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, 0x0e, 0x2e, 0x62, 0x61, 0x63, 0x6b, - 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x70, 0x6c, 0x79, 0x22, 0x00, 0x12, 0x35, 0x0a, 0x09, 0x4c, - 0x6f, 0x61, 0x64, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x12, 0x15, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, - 0x6e, 0x64, 0x2e, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x4f, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, - 0x0f, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, - 0x22, 0x00, 0x12, 0x3c, 0x0a, 0x0d, 0x50, 0x72, 0x65, 0x64, 0x69, 0x63, 0x74, 0x53, 0x74, 0x72, - 0x65, 0x61, 0x6d, 0x12, 0x17, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x50, 0x72, - 0x65, 0x64, 0x69, 0x63, 0x74, 0x4f, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, 0x0e, 0x2e, 0x62, - 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x70, 0x6c, 0x79, 0x22, 0x00, 0x30, 0x01, - 0x12, 0x40, 0x0a, 0x09, 0x45, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, 0x12, 0x17, 0x2e, - 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x50, 0x72, 0x65, 0x64, 0x69, 0x63, 0x74, 0x4f, - 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, 0x18, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, - 0x2e, 0x45, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, - 0x22, 0x00, 0x12, 0x41, 0x0a, 0x0d, 0x47, 0x65, 0x6e, 0x65, 0x72, 0x61, 0x74, 0x65, 0x49, 0x6d, - 0x61, 0x67, 0x65, 0x12, 0x1d, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x47, 0x65, - 0x6e, 0x65, 0x72, 0x61, 0x74, 0x65, 0x49, 0x6d, 0x61, 0x67, 0x65, 0x52, 0x65, 0x71, 0x75, 0x65, - 0x73, 0x74, 0x1a, 0x0f, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x73, - 0x75, 0x6c, 0x74, 0x22, 0x00, 0x12, 0x4d, 0x0a, 0x12, 0x41, 0x75, 0x64, 0x69, 0x6f, 0x54, 0x72, - 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x12, 0x1a, 0x2e, 0x62, 0x61, - 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, - 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x1a, 0x19, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, - 0x64, 0x2e, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x52, 0x65, 0x73, 0x75, - 0x6c, 0x74, 0x22, 0x00, 0x12, 0x2d, 0x0a, 0x03, 0x54, 0x54, 0x53, 0x12, 0x13, 0x2e, 0x62, 0x61, - 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x54, 0x54, 0x53, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, - 0x1a, 0x0f, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x73, 0x75, 0x6c, - 0x74, 0x22, 0x00, 0x42, 0x5a, 0x0a, 0x19, 0x69, 0x6f, 0x2e, 0x73, 0x6b, 0x79, 0x6e, 0x65, 0x74, - 0x2e, 0x6c, 0x6f, 0x63, 0x61, 0x6c, 0x61, 0x69, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, - 0x42, 0x0e, 0x4c, 0x6f, 0x63, 0x61, 0x6c, 0x41, 0x49, 0x42, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, - 0x50, 0x01, 0x5a, 0x2b, 0x67, 0x69, 0x74, 0x68, 0x75, 0x62, 0x2e, 0x63, 0x6f, 0x6d, 0x2f, 0x67, - 0x6f, 0x2d, 0x73, 0x6b, 0x79, 0x6e, 0x65, 0x74, 0x2f, 0x4c, 0x6f, 0x63, 0x61, 0x6c, 0x41, 0x49, - 0x2f, 0x70, 0x6b, 0x67, 0x2f, 0x67, 0x72, 0x70, 0x63, 0x2f, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x62, - 0x06, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x33, + 0x41, 0x18, 0x14, 0x20, 0x01, 0x28, 0x05, 0x52, 0x04, 0x4e, 0x47, 0x51, 0x41, 0x12, 0x1c, 0x0a, + 0x09, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x46, 0x69, 0x6c, 0x65, 0x18, 0x15, 0x20, 0x01, 0x28, 0x09, + 0x52, 0x09, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x46, 0x69, 0x6c, 0x65, 0x12, 0x16, 0x0a, 0x06, 0x44, + 0x65, 0x76, 0x69, 0x63, 0x65, 0x18, 0x16, 0x20, 0x01, 0x28, 0x09, 0x52, 0x06, 0x44, 0x65, 0x76, + 0x69, 0x63, 0x65, 0x12, 0x1c, 0x0a, 0x09, 0x55, 0x73, 0x65, 0x54, 0x72, 0x69, 0x74, 0x6f, 0x6e, + 0x18, 0x17, 0x20, 0x01, 0x28, 0x08, 0x52, 0x09, 0x55, 0x73, 0x65, 0x54, 0x72, 0x69, 0x74, 0x6f, + 0x6e, 0x12, 0x24, 0x0a, 0x0d, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x42, 0x61, 0x73, 0x65, 0x4e, 0x61, + 0x6d, 0x65, 0x18, 0x18, 0x20, 0x01, 0x28, 0x09, 0x52, 0x0d, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x42, + 0x61, 0x73, 0x65, 0x4e, 0x61, 0x6d, 0x65, 0x12, 0x2a, 0x0a, 0x10, 0x55, 0x73, 0x65, 0x46, 0x61, + 0x73, 0x74, 0x54, 0x6f, 0x6b, 0x65, 0x6e, 0x69, 0x7a, 0x65, 0x72, 0x18, 0x19, 0x20, 0x01, 0x28, + 0x08, 0x52, 0x10, 0x55, 0x73, 0x65, 0x46, 0x61, 0x73, 0x74, 0x54, 0x6f, 0x6b, 0x65, 0x6e, 0x69, + 0x7a, 0x65, 0x72, 0x22, 0x3c, 0x0a, 0x06, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x12, 0x18, 0x0a, + 0x07, 0x6d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x18, 0x01, 0x20, 0x01, 0x28, 0x09, 0x52, 0x07, + 0x6d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x12, 0x18, 0x0a, 0x07, 0x73, 0x75, 0x63, 0x63, 0x65, + 0x73, 0x73, 0x18, 0x02, 0x20, 0x01, 0x28, 0x08, 0x52, 0x07, 0x73, 0x75, 0x63, 0x63, 0x65, 0x73, + 0x73, 0x22, 0x31, 0x0a, 0x0f, 0x45, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, 0x52, 0x65, + 0x73, 0x75, 0x6c, 0x74, 0x12, 0x1e, 0x0a, 0x0a, 0x65, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, + 0x67, 0x73, 0x18, 0x01, 0x20, 0x03, 0x28, 0x02, 0x52, 0x0a, 0x65, 0x6d, 0x62, 0x65, 0x64, 0x64, + 0x69, 0x6e, 0x67, 0x73, 0x22, 0x5b, 0x0a, 0x11, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, + 0x70, 0x74, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x12, 0x10, 0x0a, 0x03, 0x64, 0x73, 0x74, + 0x18, 0x02, 0x20, 0x01, 0x28, 0x09, 0x52, 0x03, 0x64, 0x73, 0x74, 0x12, 0x1a, 0x0a, 0x08, 0x6c, + 0x61, 0x6e, 0x67, 0x75, 0x61, 0x67, 0x65, 0x18, 0x03, 0x20, 0x01, 0x28, 0x09, 0x52, 0x08, 0x6c, + 0x61, 0x6e, 0x67, 0x75, 0x61, 0x67, 0x65, 0x12, 0x18, 0x0a, 0x07, 0x74, 0x68, 0x72, 0x65, 0x61, + 0x64, 0x73, 0x18, 0x04, 0x20, 0x01, 0x28, 0x0d, 0x52, 0x07, 0x74, 0x68, 0x72, 0x65, 0x61, 0x64, + 0x73, 0x22, 0x5e, 0x0a, 0x10, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x52, + 0x65, 0x73, 0x75, 0x6c, 0x74, 0x12, 0x36, 0x0a, 0x08, 0x73, 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, + 0x73, 0x18, 0x01, 0x20, 0x03, 0x28, 0x0b, 0x32, 0x1a, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, + 0x64, 0x2e, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x53, 0x65, 0x67, 0x6d, + 0x65, 0x6e, 0x74, 0x52, 0x08, 0x73, 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, 0x73, 0x12, 0x12, 0x0a, + 0x04, 0x74, 0x65, 0x78, 0x74, 0x18, 0x02, 0x20, 0x01, 0x28, 0x09, 0x52, 0x04, 0x74, 0x65, 0x78, + 0x74, 0x22, 0x77, 0x0a, 0x11, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x53, + 0x65, 0x67, 0x6d, 0x65, 0x6e, 0x74, 0x12, 0x0e, 0x0a, 0x02, 0x69, 0x64, 0x18, 0x01, 0x20, 0x01, + 0x28, 0x05, 0x52, 0x02, 0x69, 0x64, 0x12, 0x14, 0x0a, 0x05, 0x73, 0x74, 0x61, 0x72, 0x74, 0x18, + 0x02, 0x20, 0x01, 0x28, 0x03, 0x52, 0x05, 0x73, 0x74, 0x61, 0x72, 0x74, 0x12, 0x10, 0x0a, 0x03, + 0x65, 0x6e, 0x64, 0x18, 0x03, 0x20, 0x01, 0x28, 0x03, 0x52, 0x03, 0x65, 0x6e, 0x64, 0x12, 0x12, + 0x0a, 0x04, 0x74, 0x65, 0x78, 0x74, 0x18, 0x04, 0x20, 0x01, 0x28, 0x09, 0x52, 0x04, 0x74, 0x65, + 0x78, 0x74, 0x12, 0x16, 0x0a, 0x06, 0x74, 0x6f, 0x6b, 0x65, 0x6e, 0x73, 0x18, 0x05, 0x20, 0x03, + 0x28, 0x05, 0x52, 0x06, 0x74, 0x6f, 0x6b, 0x65, 0x6e, 0x73, 0x22, 0xe4, 0x01, 0x0a, 0x14, 0x47, + 0x65, 0x6e, 0x65, 0x72, 0x61, 0x74, 0x65, 0x49, 0x6d, 0x61, 0x67, 0x65, 0x52, 0x65, 0x71, 0x75, + 0x65, 0x73, 0x74, 0x12, 0x16, 0x0a, 0x06, 0x68, 0x65, 0x69, 0x67, 0x68, 0x74, 0x18, 0x01, 0x20, + 0x01, 0x28, 0x05, 0x52, 0x06, 0x68, 0x65, 0x69, 0x67, 0x68, 0x74, 0x12, 0x14, 0x0a, 0x05, 0x77, + 0x69, 0x64, 0x74, 0x68, 0x18, 0x02, 0x20, 0x01, 0x28, 0x05, 0x52, 0x05, 0x77, 0x69, 0x64, 0x74, + 0x68, 0x12, 0x12, 0x0a, 0x04, 0x6d, 0x6f, 0x64, 0x65, 0x18, 0x03, 0x20, 0x01, 0x28, 0x05, 0x52, + 0x04, 0x6d, 0x6f, 0x64, 0x65, 0x12, 0x12, 0x0a, 0x04, 0x73, 0x74, 0x65, 0x70, 0x18, 0x04, 0x20, + 0x01, 0x28, 0x05, 0x52, 0x04, 0x73, 0x74, 0x65, 0x70, 0x12, 0x12, 0x0a, 0x04, 0x73, 0x65, 0x65, + 0x64, 0x18, 0x05, 0x20, 0x01, 0x28, 0x05, 0x52, 0x04, 0x73, 0x65, 0x65, 0x64, 0x12, 0x27, 0x0a, + 0x0f, 0x70, 0x6f, 0x73, 0x69, 0x74, 0x69, 0x76, 0x65, 0x5f, 0x70, 0x72, 0x6f, 0x6d, 0x70, 0x74, + 0x18, 0x06, 0x20, 0x01, 0x28, 0x09, 0x52, 0x0e, 0x70, 0x6f, 0x73, 0x69, 0x74, 0x69, 0x76, 0x65, + 0x50, 0x72, 0x6f, 0x6d, 0x70, 0x74, 0x12, 0x27, 0x0a, 0x0f, 0x6e, 0x65, 0x67, 0x61, 0x74, 0x69, + 0x76, 0x65, 0x5f, 0x70, 0x72, 0x6f, 0x6d, 0x70, 0x74, 0x18, 0x07, 0x20, 0x01, 0x28, 0x09, 0x52, + 0x0e, 0x6e, 0x65, 0x67, 0x61, 0x74, 0x69, 0x76, 0x65, 0x50, 0x72, 0x6f, 0x6d, 0x70, 0x74, 0x12, + 0x10, 0x0a, 0x03, 0x64, 0x73, 0x74, 0x18, 0x08, 0x20, 0x01, 0x28, 0x09, 0x52, 0x03, 0x64, 0x73, + 0x74, 0x22, 0x48, 0x0a, 0x0a, 0x54, 0x54, 0x53, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x12, + 0x12, 0x0a, 0x04, 0x74, 0x65, 0x78, 0x74, 0x18, 0x01, 0x20, 0x01, 0x28, 0x09, 0x52, 0x04, 0x74, + 0x65, 0x78, 0x74, 0x12, 0x14, 0x0a, 0x05, 0x6d, 0x6f, 0x64, 0x65, 0x6c, 0x18, 0x02, 0x20, 0x01, + 0x28, 0x09, 0x52, 0x05, 0x6d, 0x6f, 0x64, 0x65, 0x6c, 0x12, 0x10, 0x0a, 0x03, 0x64, 0x73, 0x74, + 0x18, 0x03, 0x20, 0x01, 0x28, 0x09, 0x52, 0x03, 0x64, 0x73, 0x74, 0x32, 0xeb, 0x03, 0x0a, 0x07, + 0x42, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x12, 0x32, 0x0a, 0x06, 0x48, 0x65, 0x61, 0x6c, 0x74, + 0x68, 0x12, 0x16, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x48, 0x65, 0x61, 0x6c, + 0x74, 0x68, 0x4d, 0x65, 0x73, 0x73, 0x61, 0x67, 0x65, 0x1a, 0x0e, 0x2e, 0x62, 0x61, 0x63, 0x6b, + 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x70, 0x6c, 0x79, 0x22, 0x00, 0x12, 0x34, 0x0a, 0x07, 0x50, + 0x72, 0x65, 0x64, 0x69, 0x63, 0x74, 0x12, 0x17, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, + 0x2e, 0x50, 0x72, 0x65, 0x64, 0x69, 0x63, 0x74, 0x4f, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, + 0x0e, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x70, 0x6c, 0x79, 0x22, + 0x00, 0x12, 0x35, 0x0a, 0x09, 0x4c, 0x6f, 0x61, 0x64, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x12, 0x15, + 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x4f, 0x70, + 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, 0x0f, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, + 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x22, 0x00, 0x12, 0x3c, 0x0a, 0x0d, 0x50, 0x72, 0x65, 0x64, + 0x69, 0x63, 0x74, 0x53, 0x74, 0x72, 0x65, 0x61, 0x6d, 0x12, 0x17, 0x2e, 0x62, 0x61, 0x63, 0x6b, + 0x65, 0x6e, 0x64, 0x2e, 0x50, 0x72, 0x65, 0x64, 0x69, 0x63, 0x74, 0x4f, 0x70, 0x74, 0x69, 0x6f, + 0x6e, 0x73, 0x1a, 0x0e, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x70, + 0x6c, 0x79, 0x22, 0x00, 0x30, 0x01, 0x12, 0x40, 0x0a, 0x09, 0x45, 0x6d, 0x62, 0x65, 0x64, 0x64, + 0x69, 0x6e, 0x67, 0x12, 0x17, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x50, 0x72, + 0x65, 0x64, 0x69, 0x63, 0x74, 0x4f, 0x70, 0x74, 0x69, 0x6f, 0x6e, 0x73, 0x1a, 0x18, 0x2e, 0x62, + 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x45, 0x6d, 0x62, 0x65, 0x64, 0x64, 0x69, 0x6e, 0x67, + 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x22, 0x00, 0x12, 0x41, 0x0a, 0x0d, 0x47, 0x65, 0x6e, 0x65, + 0x72, 0x61, 0x74, 0x65, 0x49, 0x6d, 0x61, 0x67, 0x65, 0x12, 0x1d, 0x2e, 0x62, 0x61, 0x63, 0x6b, + 0x65, 0x6e, 0x64, 0x2e, 0x47, 0x65, 0x6e, 0x65, 0x72, 0x61, 0x74, 0x65, 0x49, 0x6d, 0x61, 0x67, + 0x65, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x1a, 0x0f, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, + 0x6e, 0x64, 0x2e, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x22, 0x00, 0x12, 0x4d, 0x0a, 0x12, 0x41, + 0x75, 0x64, 0x69, 0x6f, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x69, 0x6f, + 0x6e, 0x12, 0x1a, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x54, 0x72, 0x61, 0x6e, + 0x73, 0x63, 0x72, 0x69, 0x70, 0x74, 0x52, 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x1a, 0x19, 0x2e, + 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x54, 0x72, 0x61, 0x6e, 0x73, 0x63, 0x72, 0x69, + 0x70, 0x74, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x22, 0x00, 0x12, 0x2d, 0x0a, 0x03, 0x54, 0x54, + 0x53, 0x12, 0x13, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x2e, 0x54, 0x54, 0x53, 0x52, + 0x65, 0x71, 0x75, 0x65, 0x73, 0x74, 0x1a, 0x0f, 0x2e, 0x62, 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, + 0x2e, 0x52, 0x65, 0x73, 0x75, 0x6c, 0x74, 0x22, 0x00, 0x42, 0x5a, 0x0a, 0x19, 0x69, 0x6f, 0x2e, + 0x73, 0x6b, 0x79, 0x6e, 0x65, 0x74, 0x2e, 0x6c, 0x6f, 0x63, 0x61, 0x6c, 0x61, 0x69, 0x2e, 0x62, + 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x42, 0x0e, 0x4c, 0x6f, 0x63, 0x61, 0x6c, 0x41, 0x49, 0x42, + 0x61, 0x63, 0x6b, 0x65, 0x6e, 0x64, 0x50, 0x01, 0x5a, 0x2b, 0x67, 0x69, 0x74, 0x68, 0x75, 0x62, + 0x2e, 0x63, 0x6f, 0x6d, 0x2f, 0x67, 0x6f, 0x2d, 0x73, 0x6b, 0x79, 0x6e, 0x65, 0x74, 0x2f, 0x4c, + 0x6f, 0x63, 0x61, 0x6c, 0x41, 0x49, 0x2f, 0x70, 0x6b, 0x67, 0x2f, 0x67, 0x72, 0x70, 0x63, 0x2f, + 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x62, 0x06, 0x70, 0x72, 0x6f, 0x74, 0x6f, 0x33, } var ( diff --git a/pkg/grpc/proto/backend.proto b/pkg/grpc/proto/backend.proto index c17a7644b..1945d99a9 100644 --- a/pkg/grpc/proto/backend.proto +++ b/pkg/grpc/proto/backend.proto @@ -89,6 +89,13 @@ message ModelOptions { float RopeFreqScale = 18; float RMSNormEps = 19; int32 NGQA = 20; + string ModelFile = 21; + + // AutoGPTQ + string Device = 22; + bool UseTriton = 23; + string ModelBaseName = 24; + bool UseFastTokenizer = 25; } message Result { diff --git a/pkg/grpc/transcribe/whisper.go b/pkg/grpc/transcribe/whisper.go index c0120dbda..3d25b050e 100644 --- a/pkg/grpc/transcribe/whisper.go +++ b/pkg/grpc/transcribe/whisper.go @@ -17,7 +17,7 @@ type Whisper struct { func (sd *Whisper) Load(opts *pb.ModelOptions) error { // Note: the Model here is a path to a directory containing the model files - w, err := whisper.New(opts.Model) + w, err := whisper.New(opts.ModelFile) sd.whisper = w return err } diff --git a/pkg/grpc/tts/piper.go b/pkg/grpc/tts/piper.go index 3bc85e0d4..e5e99fc80 100644 --- a/pkg/grpc/tts/piper.go +++ b/pkg/grpc/tts/piper.go @@ -18,8 +18,8 @@ type Piper struct { } func (sd *Piper) Load(opts *pb.ModelOptions) error { - if filepath.Ext(opts.Model) != ".onnx" { - return fmt.Errorf("unsupported model type %s (should end with .onnx)", opts.Model) + if filepath.Ext(opts.ModelFile) != ".onnx" { + return fmt.Errorf("unsupported model type %s (should end with .onnx)", opts.ModelFile) } var err error // Note: the Model here is a path to a directory containing the model files diff --git a/pkg/model/initializers.go b/pkg/model/initializers.go index 75644b285..2e011a3c1 100644 --- a/pkg/model/initializers.go +++ b/pkg/model/initializers.go @@ -83,7 +83,9 @@ func (ml *ModelLoader) startProcess(grpcProcess, id string, serverAddress string grpcControlProcess := process.New( process.WithTemporaryStateDir(), process.WithName(grpcProcess), - process.WithArgs("--addr", serverAddress)) + process.WithArgs("--addr", serverAddress), + process.WithEnvironment(os.Environ()...), + ) ml.grpcProcesses[id] = grpcControlProcess @@ -124,8 +126,8 @@ func (ml *ModelLoader) startProcess(grpcProcess, id string, serverAddress string // starts the grpcModelProcess for the backend, and returns a grpc client // It also loads the model -func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string) (*grpc.Client, error) { - return func(s string) (*grpc.Client, error) { +func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string, string) (*grpc.Client, error) { + return func(modelName, modelFile string) (*grpc.Client, error) { log.Debug().Msgf("Loading GRPC Model %s: %+v", backend, *o) var client *grpc.Client @@ -148,7 +150,7 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string) (*grpc return nil, fmt.Errorf("failed allocating free ports: %s", err.Error()) } // Make sure the process is executable - if err := ml.startProcess(uri, o.modelFile, serverAddress); err != nil { + if err := ml.startProcess(uri, o.model, serverAddress); err != nil { return nil, err } @@ -172,7 +174,7 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string) (*grpc } // Make sure the process is executable - if err := ml.startProcess(grpcProcess, o.modelFile, serverAddress); err != nil { + if err := ml.startProcess(grpcProcess, o.model, serverAddress); err != nil { return nil, err } @@ -198,7 +200,8 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string) (*grpc } options := *o.gRPCOptions - options.Model = s + options.Model = modelName + options.ModelFile = modelFile log.Debug().Msgf("GRPC: Loading model with options: %+v", options) @@ -217,14 +220,14 @@ func (ml *ModelLoader) grpcModel(backend string, o *Options) func(string) (*grpc func (ml *ModelLoader) BackendLoader(opts ...Option) (model *grpc.Client, err error) { o := NewOptions(opts...) - log.Debug().Msgf("Loading model %s from %s", o.backendString, o.modelFile) + log.Debug().Msgf("Loading model %s from %s", o.backendString, o.model) backend := strings.ToLower(o.backendString) // if an external backend is provided, use it _, externalBackendExists := o.externalBackends[backend] if externalBackendExists { - return ml.LoadModel(o.modelFile, ml.grpcModel(backend, o)) + return ml.LoadModel(o.model, ml.grpcModel(backend, o)) } switch backend { @@ -232,13 +235,13 @@ func (ml *ModelLoader) BackendLoader(opts ...Option) (model *grpc.Client, err er MPTBackend, Gpt2Backend, FalconBackend, GPTNeoXBackend, ReplitBackend, StarcoderBackend, BloomzBackend, RwkvBackend, LCHuggingFaceBackend, BertEmbeddingsBackend, FalconGGMLBackend, StableDiffusionBackend, WhisperBackend: - return ml.LoadModel(o.modelFile, ml.grpcModel(backend, o)) + return ml.LoadModel(o.model, ml.grpcModel(backend, o)) case Gpt4AllLlamaBackend, Gpt4AllMptBackend, Gpt4AllJBackend, Gpt4All: o.gRPCOptions.LibrarySearchPath = filepath.Join(o.assetDir, "backend-assets", "gpt4all") - return ml.LoadModel(o.modelFile, ml.grpcModel(Gpt4All, o)) + return ml.LoadModel(o.model, ml.grpcModel(Gpt4All, o)) case PiperBackend: o.gRPCOptions.LibrarySearchPath = filepath.Join(o.assetDir, "backend-assets", "espeak-ng-data") - return ml.LoadModel(o.modelFile, ml.grpcModel(PiperBackend, o)) + return ml.LoadModel(o.model, ml.grpcModel(PiperBackend, o)) default: return nil, fmt.Errorf("backend unsupported: %s", o.backendString) } @@ -249,8 +252,8 @@ func (ml *ModelLoader) GreedyLoader(opts ...Option) (*grpc.Client, error) { // Is this really needed? BackendLoader already does this ml.mu.Lock() - if m := ml.checkIsLoaded(o.modelFile); m != nil { - log.Debug().Msgf("Model '%s' already loaded", o.modelFile) + if m := ml.checkIsLoaded(o.model); m != nil { + log.Debug().Msgf("Model '%s' already loaded", o.model) ml.mu.Unlock() return m, nil } @@ -263,13 +266,13 @@ func (ml *ModelLoader) GreedyLoader(opts ...Option) (*grpc.Client, error) { for _, b := range o.externalBackends { allBackendsToAutoLoad = append(allBackendsToAutoLoad, b) } - log.Debug().Msgf("Loading model '%s' greedly from all the available backends: %s", o.modelFile, strings.Join(allBackendsToAutoLoad, ", ")) + log.Debug().Msgf("Loading model '%s' greedly from all the available backends: %s", o.model, strings.Join(allBackendsToAutoLoad, ", ")) for _, b := range allBackendsToAutoLoad { log.Debug().Msgf("[%s] Attempting to load", b) options := []Option{ WithBackendString(b), - WithModelFile(o.modelFile), + WithModel(o.model), WithLoadGRPCLLMModelOpts(o.gRPCOptions), WithThreads(o.threads), WithAssetDir(o.assetDir), diff --git a/pkg/model/loader.go b/pkg/model/loader.go index a4d86dbce..b45a52c85 100644 --- a/pkg/model/loader.go +++ b/pkg/model/loader.go @@ -98,7 +98,7 @@ func (ml *ModelLoader) ListModels() ([]string, error) { return models, nil } -func (ml *ModelLoader) LoadModel(modelName string, loader func(string) (*grpc.Client, error)) (*grpc.Client, error) { +func (ml *ModelLoader) LoadModel(modelName string, loader func(string, string) (*grpc.Client, error)) (*grpc.Client, error) { ml.mu.Lock() defer ml.mu.Unlock() @@ -111,7 +111,7 @@ func (ml *ModelLoader) LoadModel(modelName string, loader func(string) (*grpc.Cl modelFile := filepath.Join(ml.ModelPath, modelName) log.Debug().Msgf("Loading model in memory from file: %s", modelFile) - model, err := loader(modelFile) + model, err := loader(modelName, modelFile) if err != nil { return nil, err } diff --git a/pkg/model/options.go b/pkg/model/options.go index 466e9c2f5..646a52237 100644 --- a/pkg/model/options.go +++ b/pkg/model/options.go @@ -8,7 +8,7 @@ import ( type Options struct { backendString string - modelFile string + model string threads uint32 assetDir string context context.Context @@ -35,9 +35,9 @@ func WithBackendString(backend string) Option { } } -func WithModelFile(modelFile string) Option { +func WithModel(modelFile string) Option { return func(o *Options) { - o.modelFile = modelFile + o.model = modelFile } }