# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project # mypy: ignore-errors # -*- coding: utf-8 -*- # Generated by the protocol buffer compiler. DO NOT EDIT! # NO CHECKED-IN PROTOBUF GENCODE # source: vllm/grpc/vllm_engine.proto # Protobuf Python Version: 6.31.1 """Generated protocol buffer code.""" from google.protobuf import descriptor as _descriptor from google.protobuf import descriptor_pool as _descriptor_pool from google.protobuf import runtime_version as _runtime_version from google.protobuf import symbol_database as _symbol_database from google.protobuf.internal import builder as _builder _runtime_version.ValidateProtobufRuntimeVersion( _runtime_version.Domain.PUBLIC, 6, 31, 1, '', 'vllm/grpc/vllm_engine.proto' ) # @@protoc_insertion_point(imports) _sym_db = _symbol_database.Default() DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x1bvllm/grpc/vllm_engine.proto\x12\x10vllm.grpc.engine\"\xfa\x06\n\x0eSamplingParams\x12\x18\n\x0btemperature\x18\x01 \x01(\x02H\x01\x88\x01\x01\x12\r\n\x05top_p\x18\x02 \x01(\x02\x12\r\n\x05top_k\x18\x03 \x01(\r\x12\r\n\x05min_p\x18\x04 \x01(\x02\x12\x19\n\x11\x66requency_penalty\x18\x05 \x01(\x02\x12\x18\n\x10presence_penalty\x18\x06 \x01(\x02\x12\x1a\n\x12repetition_penalty\x18\x07 \x01(\x02\x12\x17\n\nmax_tokens\x18\x08 \x01(\rH\x02\x88\x01\x01\x12\x12\n\nmin_tokens\x18\t \x01(\r\x12\x0c\n\x04stop\x18\n \x03(\t\x12\x16\n\x0estop_token_ids\x18\x0b \x03(\r\x12\x1b\n\x13skip_special_tokens\x18\x0c \x01(\x08\x12%\n\x1dspaces_between_special_tokens\x18\r \x01(\x08\x12\x12\n\nignore_eos\x18\x0e \x01(\x08\x12\t\n\x01n\x18\x0f \x01(\r\x12\x15\n\x08logprobs\x18\x16 \x01(\x05H\x03\x88\x01\x01\x12\x1c\n\x0fprompt_logprobs\x18\x17 \x01(\x05H\x04\x88\x01\x01\x12\x11\n\x04seed\x18\x18 \x01(\x05H\x05\x88\x01\x01\x12\"\n\x1ainclude_stop_str_in_output\x18\x19 \x01(\x08\x12\x43\n\nlogit_bias\x18\x1a \x03(\x0b\x32/.vllm.grpc.engine.SamplingParams.LogitBiasEntry\x12#\n\x16truncate_prompt_tokens\x18\x1b \x01(\x05H\x06\x88\x01\x01\x12\x15\n\x0bjson_schema\x18\x10 \x01(\tH\x00\x12\x0f\n\x05regex\x18\x11 \x01(\tH\x00\x12\x11\n\x07grammar\x18\x12 \x01(\tH\x00\x12\x18\n\x0estructural_tag\x18\x13 \x01(\tH\x00\x12\x15\n\x0bjson_object\x18\x14 \x01(\x08H\x00\x12\x34\n\x06\x63hoice\x18\x15 \x01(\x0b\x32\".vllm.grpc.engine.ChoiceConstraintH\x00\x1a\x30\n\x0eLogitBiasEntry\x12\x0b\n\x03key\x18\x01 \x01(\x05\x12\r\n\x05value\x18\x02 \x01(\x02:\x02\x38\x01\x42\x0c\n\nconstraintB\x0e\n\x0c_temperatureB\r\n\x0b_max_tokensB\x0b\n\t_logprobsB\x12\n\x10_prompt_logprobsB\x07\n\x05_seedB\x19\n\x17_truncate_prompt_tokens\"#\n\x10\x43hoiceConstraint\x12\x0f\n\x07\x63hoices\x18\x01 \x03(\t\":\n\x0eTokenizedInput\x12\x15\n\roriginal_text\x18\x01 \x01(\t\x12\x11\n\tinput_ids\x18\x02 \x03(\r\"\xc0\x01\n\x0fGenerateRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x35\n\ttokenized\x18\x02 \x01(\x0b\x32 .vllm.grpc.engine.TokenizedInputH\x00\x12\x0e\n\x04text\x18\x03 \x01(\tH\x00\x12\x39\n\x0fsampling_params\x18\x04 \x01(\x0b\x32 .vllm.grpc.engine.SamplingParams\x12\x0e\n\x06stream\x18\x05 \x01(\x08\x42\x07\n\x05input\"\x8e\x01\n\x10GenerateResponse\x12\x36\n\x05\x63hunk\x18\x01 \x01(\x0b\x32%.vllm.grpc.engine.GenerateStreamChunkH\x00\x12\x36\n\x08\x63omplete\x18\x02 \x01(\x0b\x32\".vllm.grpc.engine.GenerateCompleteH\x00\x42\n\n\x08response\"q\n\x13GenerateStreamChunk\x12\x11\n\ttoken_ids\x18\x01 \x03(\r\x12\x15\n\rprompt_tokens\x18\x02 \x01(\r\x12\x19\n\x11\x63ompletion_tokens\x18\x03 \x01(\r\x12\x15\n\rcached_tokens\x18\x04 \x01(\r\"\x86\x01\n\x10GenerateComplete\x12\x12\n\noutput_ids\x18\x01 \x03(\r\x12\x15\n\rfinish_reason\x18\x02 \x01(\t\x12\x15\n\rprompt_tokens\x18\x03 \x01(\r\x12\x19\n\x11\x63ompletion_tokens\x18\x04 \x01(\r\x12\x15\n\rcached_tokens\x18\x05 \x01(\r\"W\n\x0c\x45mbedRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x33\n\ttokenized\x18\x02 \x01(\x0b\x32 .vllm.grpc.engine.TokenizedInput\"P\n\rEmbedResponse\x12\x11\n\tembedding\x18\x01 \x03(\x02\x12\x15\n\rprompt_tokens\x18\x02 \x01(\r\x12\x15\n\rembedding_dim\x18\x03 \x01(\r\"\x14\n\x12HealthCheckRequest\"7\n\x13HealthCheckResponse\x12\x0f\n\x07healthy\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"#\n\x0c\x41\x62ortRequest\x12\x13\n\x0brequest_ids\x18\x01 \x03(\t\"\x0f\n\rAbortResponse\"\x15\n\x13GetModelInfoRequest\"\x8a\x01\n\x14GetModelInfoResponse\x12\x12\n\nmodel_path\x18\x01 \x01(\t\x12\x15\n\ris_generation\x18\x02 \x01(\x08\x12\x1a\n\x12max_context_length\x18\x03 \x01(\r\x12\x12\n\nvocab_size\x18\x04 \x01(\r\x12\x17\n\x0fsupports_vision\x18\x05 \x01(\x08\"\x16\n\x14GetServerInfoRequest\"\x90\x01\n\x15GetServerInfoResponse\x12\x17\n\x0f\x61\x63tive_requests\x18\x01 \x01(\r\x12\x11\n\tis_paused\x18\x02 \x01(\x08\x12\x1e\n\x16last_receive_timestamp\x18\x03 \x01(\x01\x12\x16\n\x0euptime_seconds\x18\x04 \x01(\x01\x12\x13\n\x0bserver_type\x18\x05 \x01(\t2\x92\x04\n\nVllmEngine\x12S\n\x08Generate\x12!.vllm.grpc.engine.GenerateRequest\x1a\".vllm.grpc.engine.GenerateResponse0\x01\x12H\n\x05\x45mbed\x12\x1e.vllm.grpc.engine.EmbedRequest\x1a\x1f.vllm.grpc.engine.EmbedResponse\x12Z\n\x0bHealthCheck\x12$.vllm.grpc.engine.HealthCheckRequest\x1a%.vllm.grpc.engine.HealthCheckResponse\x12H\n\x05\x41\x62ort\x12\x1e.vllm.grpc.engine.AbortRequest\x1a\x1f.vllm.grpc.engine.AbortResponse\x12]\n\x0cGetModelInfo\x12%.vllm.grpc.engine.GetModelInfoRequest\x1a&.vllm.grpc.engine.GetModelInfoResponse\x12`\n\rGetServerInfo\x12&.vllm.grpc.engine.GetServerInfoRequest\x1a\'.vllm.grpc.engine.GetServerInfoResponseb\x06proto3') _globals = globals() _builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals) _builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'vllm.grpc.vllm_engine_pb2', _globals) if not _descriptor._USE_C_DESCRIPTORS: DESCRIPTOR._loaded_options = None _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._loaded_options = None _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_options = b'8\001' _globals['_SAMPLINGPARAMS']._serialized_start=50 _globals['_SAMPLINGPARAMS']._serialized_end=940 _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_start=778 _globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_end=826 _globals['_CHOICECONSTRAINT']._serialized_start=942 _globals['_CHOICECONSTRAINT']._serialized_end=977 _globals['_TOKENIZEDINPUT']._serialized_start=979 _globals['_TOKENIZEDINPUT']._serialized_end=1037 _globals['_GENERATEREQUEST']._serialized_start=1040 _globals['_GENERATEREQUEST']._serialized_end=1232 _globals['_GENERATERESPONSE']._serialized_start=1235 _globals['_GENERATERESPONSE']._serialized_end=1377 _globals['_GENERATESTREAMCHUNK']._serialized_start=1379 _globals['_GENERATESTREAMCHUNK']._serialized_end=1492 _globals['_GENERATECOMPLETE']._serialized_start=1495 _globals['_GENERATECOMPLETE']._serialized_end=1629 _globals['_EMBEDREQUEST']._serialized_start=1631 _globals['_EMBEDREQUEST']._serialized_end=1718 _globals['_EMBEDRESPONSE']._serialized_start=1720 _globals['_EMBEDRESPONSE']._serialized_end=1800 _globals['_HEALTHCHECKREQUEST']._serialized_start=1802 _globals['_HEALTHCHECKREQUEST']._serialized_end=1822 _globals['_HEALTHCHECKRESPONSE']._serialized_start=1824 _globals['_HEALTHCHECKRESPONSE']._serialized_end=1879 _globals['_ABORTREQUEST']._serialized_start=1881 _globals['_ABORTREQUEST']._serialized_end=1916 _globals['_ABORTRESPONSE']._serialized_start=1918 _globals['_ABORTRESPONSE']._serialized_end=1933 _globals['_GETMODELINFOREQUEST']._serialized_start=1935 _globals['_GETMODELINFOREQUEST']._serialized_end=1956 _globals['_GETMODELINFORESPONSE']._serialized_start=1959 _globals['_GETMODELINFORESPONSE']._serialized_end=2097 _globals['_GETSERVERINFOREQUEST']._serialized_start=2099 _globals['_GETSERVERINFOREQUEST']._serialized_end=2121 _globals['_GETSERVERINFORESPONSE']._serialized_start=2124 _globals['_GETSERVERINFORESPONSE']._serialized_end=2268 _globals['_VLLMENGINE']._serialized_start=2271 _globals['_VLLMENGINE']._serialized_end=2801 # @@protoc_insertion_point(module_scope)