diff --git a/extra/grpc/sam/Makefile b/extra/grpc/sam/Makefile new file mode 100644 index 000000000000..68dac7731d5d --- /dev/null +++ b/extra/grpc/sam/Makefile @@ -0,0 +1,11 @@ +.PONY: sam +sam: + @echo "Creating virtual environment..." + @conda env create --name sam --file sam.yml + @echo "Virtual environment created." + +.PONY: run +run: + @echo "Running sam..." + bash run.sh + @echo "sam run." \ No newline at end of file diff --git a/extra/grpc/sam/README.md b/extra/grpc/sam/README.md new file mode 100644 index 000000000000..2ad723ac9570 --- /dev/null +++ b/extra/grpc/sam/README.md @@ -0,0 +1,5 @@ +# Creating a separate environment for the sam project + +``` +make sam +``` \ No newline at end of file diff --git a/extra/grpc/sam/backend_pb2.py b/extra/grpc/sam/backend_pb2.py new file mode 100644 index 000000000000..12e8bf51e157 --- /dev/null +++ b/extra/grpc/sam/backend_pb2.py @@ -0,0 +1,61 @@ +# -*- 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\"\x96\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\x12\x0e\n\x06NDraft\x18) \x01(\x05\"\x18\n\x05Reply\x12\x0f\n\x07message\x18\x01 \x01(\x0c\"\x86\x06\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\x12\x14\n\x0cPipelineType\x18\x1a \x01(\t\x12\x15\n\rSchedulerType\x18\x1b \x01(\t\x12\x0c\n\x04\x43UDA\x18\x1c \x01(\x08\x12\x10\n\x08\x43\x46GScale\x18\x1d \x01(\x02\x12\x0f\n\x07IMG2IMG\x18\x1e \x01(\x08\x12\x11\n\tCLIPModel\x18\x1f \x01(\t\x12\x15\n\rCLIPSubfolder\x18 \x01(\t\x12\x10\n\x08\x43LIPSkip\x18! \x01(\x05\x12\x11\n\tTokenizer\x18\" \x01(\t\x12\x10\n\x08LoraBase\x18# \x01(\t\x12\x13\n\x0bLoraAdapter\x18$ \x01(\t\x12\x11\n\tNoMulMatQ\x18% \x01(\x08\x12\x12\n\nDraftModel\x18\' \x01(\t\x12\x11\n\tAudioPath\x18& \x01(\t\x12\x14\n\x0cQuantization\x18( \x01(\t\"*\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\"\xd7\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\x12\x0b\n\x03src\x18\t \x01(\t\x12\x18\n\x10\x45nableParameters\x18\n \x01(\t\x12\x10\n\x08\x43LIPSkip\x18\x0b \x01(\x05\"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(\t\"6\n\x14TokenizationResponse\x12\x0e\n\x06length\x18\x01 \x01(\x05\x12\x0e\n\x06tokens\x18\x02 \x03(\x05\"\x8e\x01\n\x0fMemoryUsageData\x12\r\n\x05total\x18\x01 \x01(\x04\x12:\n\tbreakdown\x18\x02 \x03(\x0b\x32\'.backend.MemoryUsageData.BreakdownEntry\x1a\x30\n\x0e\x42reakdownEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x04:\x02\x38\x01\"\xad\x01\n\x0eStatusResponse\x12,\n\x05state\x18\x01 \x01(\x0e\x32\x1d.backend.StatusResponse.State\x12(\n\x06memory\x18\x02 \x01(\x0b\x32\x18.backend.MemoryUsageData\"C\n\x05State\x12\x11\n\rUNINITIALIZED\x10\x00\x12\x08\n\x04\x42USY\x10\x01\x12\t\n\x05READY\x10\x02\x12\x12\n\x05\x45RROR\x10\xff\xff\xff\xff\xff\xff\xff\xff\xff\x01\x32\xf4\x04\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\x12J\n\x0eTokenizeString\x12\x17.backend.PredictOptions\x1a\x1d.backend.TokenizationResponse\"\x00\x12;\n\x06Status\x12\x16.backend.HealthMessage\x1a\x17.backend.StatusResponse\"\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' + _MEMORYUSAGEDATA_BREAKDOWNENTRY._options = None + _MEMORYUSAGEDATA_BREAKDOWNENTRY._serialized_options = b'8\001' + _globals['_HEALTHMESSAGE']._serialized_start=26 + _globals['_HEALTHMESSAGE']._serialized_end=41 + _globals['_PREDICTOPTIONS']._serialized_start=44 + _globals['_PREDICTOPTIONS']._serialized_end=834 + _globals['_REPLY']._serialized_start=836 + _globals['_REPLY']._serialized_end=860 + _globals['_MODELOPTIONS']._serialized_start=863 + _globals['_MODELOPTIONS']._serialized_end=1637 + _globals['_RESULT']._serialized_start=1639 + _globals['_RESULT']._serialized_end=1681 + _globals['_EMBEDDINGRESULT']._serialized_start=1683 + _globals['_EMBEDDINGRESULT']._serialized_end=1720 + _globals['_TRANSCRIPTREQUEST']._serialized_start=1722 + _globals['_TRANSCRIPTREQUEST']._serialized_end=1789 + _globals['_TRANSCRIPTRESULT']._serialized_start=1791 + _globals['_TRANSCRIPTRESULT']._serialized_end=1869 + _globals['_TRANSCRIPTSEGMENT']._serialized_start=1871 + _globals['_TRANSCRIPTSEGMENT']._serialized_end=1960 + _globals['_GENERATEIMAGEREQUEST']._serialized_start=1963 + _globals['_GENERATEIMAGEREQUEST']._serialized_end=2178 + _globals['_TTSREQUEST']._serialized_start=2180 + _globals['_TTSREQUEST']._serialized_end=2234 + _globals['_TOKENIZATIONRESPONSE']._serialized_start=2236 + _globals['_TOKENIZATIONRESPONSE']._serialized_end=2290 + _globals['_MEMORYUSAGEDATA']._serialized_start=2293 + _globals['_MEMORYUSAGEDATA']._serialized_end=2435 + _globals['_MEMORYUSAGEDATA_BREAKDOWNENTRY']._serialized_start=2387 + _globals['_MEMORYUSAGEDATA_BREAKDOWNENTRY']._serialized_end=2435 + _globals['_STATUSRESPONSE']._serialized_start=2438 + _globals['_STATUSRESPONSE']._serialized_end=2611 + _globals['_STATUSRESPONSE_STATE']._serialized_start=2544 + _globals['_STATUSRESPONSE_STATE']._serialized_end=2611 + _globals['_BACKEND']._serialized_start=2614 + _globals['_BACKEND']._serialized_end=3242 +# @@protoc_insertion_point(module_scope) diff --git a/extra/grpc/sam/backend_pb2_grpc.py b/extra/grpc/sam/backend_pb2_grpc.py new file mode 100644 index 000000000000..79a7677fb27f --- /dev/null +++ b/extra/grpc/sam/backend_pb2_grpc.py @@ -0,0 +1,363 @@ +# 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, + ) + self.TokenizeString = channel.unary_unary( + '/backend.Backend/TokenizeString', + request_serializer=backend__pb2.PredictOptions.SerializeToString, + response_deserializer=backend__pb2.TokenizationResponse.FromString, + ) + self.Status = channel.unary_unary( + '/backend.Backend/Status', + request_serializer=backend__pb2.HealthMessage.SerializeToString, + response_deserializer=backend__pb2.StatusResponse.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 TokenizeString(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 Status(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, + ), + 'TokenizeString': grpc.unary_unary_rpc_method_handler( + servicer.TokenizeString, + request_deserializer=backend__pb2.PredictOptions.FromString, + response_serializer=backend__pb2.TokenizationResponse.SerializeToString, + ), + 'Status': grpc.unary_unary_rpc_method_handler( + servicer.Status, + request_deserializer=backend__pb2.HealthMessage.FromString, + response_serializer=backend__pb2.StatusResponse.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) + + @staticmethod + def TokenizeString(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/TokenizeString', + backend__pb2.PredictOptions.SerializeToString, + backend__pb2.TokenizationResponse.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) + + @staticmethod + def Status(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/Status', + backend__pb2.HealthMessage.SerializeToString, + backend__pb2.StatusResponse.FromString, + options, channel_credentials, + insecure, call_credentials, compression, wait_for_ready, timeout, metadata) diff --git a/extra/grpc/sam/run.sh b/extra/grpc/sam/run.sh new file mode 100755 index 000000000000..56839bbf26a6 --- /dev/null +++ b/extra/grpc/sam/run.sh @@ -0,0 +1,13 @@ +#!/bin/bash + +## +## A bash script wrapper that runs the sam server with conda +export PATH=$PATH:/opt/conda/bin + +# Activate conda environment +source activate sam + +# get the directory where the bash script is located +DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )" + +python $DIR/sam.py $@ \ No newline at end of file diff --git a/extra/grpc/sam/sam.py b/extra/grpc/sam/sam.py new file mode 100644 index 000000000000..d58d194e50b6 --- /dev/null +++ b/extra/grpc/sam/sam.py @@ -0,0 +1,212 @@ +#! /usr/bin/env python3 +from __future__ import annotations + +from concurrent import futures +import argparse +import os +import signal +import sys +import os +import time + +import backend_pb2 +import backend_pb2_grpc + +import grpc + +import torch +from functools import partial +from segment_anything_hq import SamAutomaticMaskGenerator +from segment_anything_hq.modeling import ImageEncoderViT, MaskDecoderHQ, PromptEncoder, Sam, TwoWayTransformer +import matplotlib.pyplot as plt +import numpy as np + + +_ONE_DAY_IN_SECONDS = 60 * 60 * 24 +PROMT_EMBED_DIM=256 +IMAGE_SIZE = 1024 +VIT_PATCH_SIZE=16 + +# Enum for sam model type +class SamModelType: + default = "sam_hq_vit_h.pth" + vit_h = "sam_hq_vit_h.pth" + vit_l = "sam_hq_vit_l.pth" + vit_b = "sam_hq_vit_b.pth" + vit_tiny = "sam_hq_vit_tiny.pth" + + +# If MAX_WORKERS are specified in the environment use it, otherwise default to 1 +MAX_WORKERS = int(os.environ.get('PYTHON_GRPC_MAX_WORKERS', '1')) + + +# Implement the BackendServicer class with the service methods +class BackendServicer(backend_pb2_grpc.BackendServicer): + """ + A gRPC servicer for the backend service. + """ + + def Health(self, request, context): + return backend_pb2.Reply(message=bytes("OK", "utf-8")) + + def LoadModel(self, request, context): + try: + model_name = request.model_name + if model_name not in SamModelType.__dict__.keys(): + raise Exception(f"Model name {model_name} not found in {SamModelType.__dict__.keys()}") + model_path = request.model_path + # check the model_path is valid + if not os.path.exists(model_path): + raise Exception(f"Model path {model_path} does not exist") + + match model_name: + case SamModelType.default: + sam = _build_sam_vit_h(checkpoint=model_path) + case SamModelType.vit_h: + sam = _build_sam_vit_h(checkpoint=model_path) + case SamModelType.vit_l: + sam = _build_sam_vit_l(checkpoint=model_path) + case SamModelType.vit_b: + sam = _build_sam_vit_b(checkpoint=model_path) + case SamModelType.vit_tiny: + # TODO: Implement this + pass + case _: + raise Exception(f"Model name {model_name} not found in {SamModelType.__dict__.keys()}") + # TODO No sure if this is the right way to do it + self.model=sam + + except Exception as err: + return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}") + return backend_pb2.Result(success=True, message="Model loaded successfully") + + def GenerateImage(self, request, context): + try: + mask_generator=SamAutomaticMaskGenerator( + model=self.model, + points_per_side=32, + pred_iou_thresh=0.8, + stability_score_thresh=0.9, + crop_n_layers=1, + crop_n_points_downscale_factor=2, + min_mask_region_area=100 + ) + + masks=mask_generator.generate_mask(request.image) + + except Exception as err: + return backend_pb2.Result(success=False, message=f"Unexpected {err=}, {type(err)=}") + + + def PredictStream(self, request, context): + return super().PredictStream(request, context) + +def _constrcut_sam(encoder_embed_dim,encoder_depth,encoder_num_heads,encoder_global_attn_indexes,checkpoint=None): + image_embedding_size = IMAGE_SIZE // VIT_PATCH_SIZE + sam = Sam( + image_encoder=ImageEncoderViT( + depth=encoder_depth, + embed_dim=encoder_embed_dim, + img_size=IMAGE_SIZE, + mlp_ratio=4, + norm_layer=partial(torch.nn.LayerNorm, eps=1e-6), + num_heads=encoder_num_heads, + patch_size=VIT_PATCH_SIZE, + qkv_bias=True, + use_rel_pos=True, + global_attn_indexes=encoder_global_attn_indexes, + window_size=14, + out_chans=PROMT_EMBED_DIM, + ), + prompt_encoder=PromptEncoder( + embed_dim=PROMT_EMBED_DIM, + image_embedding_size=(image_embedding_size, image_embedding_size), + input_image_size=(IMAGE_SIZE, IMAGE_SIZE), + mask_in_chans=16, + ), + mask_decoder=MaskDecoderHQ( + num_multimask_outputs=3, + transformer=TwoWayTransformer( + depth=2, + embedding_dim=PROMT_EMBED_DIM, + mlp_dim=2048, + num_heads=8, + ), + transformer_dim=PROMT_EMBED_DIM, + iou_head_depth=3, + iou_head_hidden_dim=256, + vit_dim=encoder_embed_dim, + ), + pixel_mean=[123.675, 116.28, 103.53], + pixel_std=[58.395, 57.12, 57.375], + ) + + sam.eval() + if checkpoint is not None: + with open(checkpoint, "rb") as f: + device = "cuda" if torch.cuda.is_available() else "cpu" + state_dict = torch.load(f, map_location=device) + info = sam.load_state_dict(state_dict, strict=False) + print(info) + for n, p in sam.named_parameters(): + if 'hf_token' not in n and 'hf_mlp' not in n and 'compress_vit_feat' not in n and 'embedding_encoder' not in n and 'embedding_maskfeature' not in n: + p.requires_grad = False + + return sam + +def _build_sam_vit_h(checkpoint=None): + return _constrcut_sam(encoder_embed_dim=1280,encoder_depth=32,encoder_num_heads=16,encoder_global_attn_indexes=[7,15,23,31],checkpoint=checkpoint) + +def _build_sam_vit_l(checkpoint=None): + return _constrcut_sam(encoder_embed_dim=1024,encoder_depth=24,encoder_num_heads=16,encoder_global_attn_indexes=[5,11,17,23],checkpoint=checkpoint) + +def _build_sam_vit_b(checkpoint=None): + return _constrcut_sam(encoder_embed_dim=768,encoder_depth=12,encoder_num_heads=12,encoder_global_attn_indexes=[2,5,8,11],checkpoint=checkpoint) + +def masks_to_image(anns): + if len(anns)==0: + return + sorted_anns = sorted(anns, key=(lambda x: x['area']), reverse=True) + ax = plt.gca() + ax.set_autoscale_on(False) + + img = np.ones((sorted_anns[0]['segmentation'].shape[0], sorted_anns[0]['segmentation'].shape[1], 4)) + img[:,:,3] = 0 + for ann in sorted_anns: + m = ann['segmentation'] + color_mask = np.concatenate([np.random.random(3), [0.35]]) + img[m] = color_mask + plt.imsave('mask.png', img) + + +def serve(address): + server = grpc.server(futures.ThreadPoolExecutor(max_workers=MAX_WORKERS)) + 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/sam/sam.yml b/extra/grpc/sam/sam.yml new file mode 100644 index 000000000000..42717a0a962f --- /dev/null +++ b/extra/grpc/sam/sam.yml @@ -0,0 +1,61 @@ +name: sam +channels: + - defaults +dependencies: + - _libgcc_mutex=0.1=main + - _openmp_mutex=5.1=1_gnu + - bzip2=1.0.8=h7b6447c_0 + - ca-certificates=2023.08.22=h06a4308_0 + - ld_impl_linux-64=2.38=h1181459_1 + - libffi=3.4.4=h6a678d5_0 + - libgcc-ng=11.2.0=h1234567_1 + - libgomp=11.2.0=h1234567_1 + - libstdcxx-ng=11.2.0=h1234567_1 + - libuuid=1.41.5=h5eee18b_0 + - ncurses=6.4=h6a678d5_0 + - openssl=3.0.12=h7f8727e_0 + - pip=23.3=py311h06a4308_0 + - python=3.11.5=h955ad1f_0 + - readline=8.2=h5eee18b_0 + - setuptools=68.0.0=py311h06a4308_0 + - sqlite=3.41.2=h5eee18b_0 + - tk=8.6.12=h1ccaba5_0 + - tzdata=2023c=h04d1e81_0 + - wheel=0.41.2=py311h06a4308_0 + - xz=5.4.2=h5eee18b_0 + - zlib=1.2.13=h5eee18b_0 + - pip: + - certifi==2023.7.22 + - charset-normalizer==3.3.2 + - filelock==3.13.1 + - fsspec==2023.10.0 + - grpcio==1.59.2 + - idna==3.4 + - jinja2==3.1.2 + - markupsafe==2.1.3 + - mpmath==1.3.0 + - networkx==3.2.1 + - numpy==1.26.1 + - nvidia-cublas-cu12==12.1.3.1 + - nvidia-cuda-cupti-cu12==12.1.105 + - nvidia-cuda-nvrtc-cu12==12.1.105 + - nvidia-cuda-runtime-cu12==12.1.105 + - nvidia-cudnn-cu12==8.9.2.26 + - nvidia-cufft-cu12==11.0.2.54 + - nvidia-curand-cu12==10.3.2.106 + - nvidia-cusolver-cu12==11.4.5.107 + - nvidia-cusparse-cu12==12.1.0.106 + - nvidia-nccl-cu12==2.18.1 + - nvidia-nvjitlink-cu12==12.3.52 + - nvidia-nvtx-cu12==12.1.105 + - pillow==10.1.0 + - protobuf==4.25.0 + - requests==2.31.0 + - segment-anything-hq==0.3 + - sympy==1.12 + - torch==2.1.0 + - torchvision==0.16.0 + - triton==2.1.0 + - typing-extensions==4.8.0 + - urllib3==2.0.7 +prefix: /opt/conda/envs/sam diff --git a/extra/requirements.txt b/extra/requirements.txt deleted file mode 100644 index fb3cc0122a9b..000000000000 --- a/extra/requirements.txt +++ /dev/null @@ -1,7 +0,0 @@ -sentence_transformers -grpcio -google -protobuf -six -omegaconf -compel \ No newline at end of file