112 lines
14 KiB
Python
112 lines
14 KiB
Python
# This file is auto-generated. Do not edit manually.
|
|
# Regenerate with: python compile_proto.py
|
|
|
|
# -*- coding: utf-8 -*-
|
|
# Generated by the protocol buffer compiler. DO NOT EDIT!
|
|
# NO CHECKED-IN PROTOBUF GENCODE
|
|
# source: sglang_scheduler.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,
|
|
'',
|
|
'sglang_scheduler.proto'
|
|
)
|
|
# @@protoc_insertion_point(imports)
|
|
|
|
_sym_db = _symbol_database.Default()
|
|
|
|
|
|
from google.protobuf import timestamp_pb2 as google_dot_protobuf_dot_timestamp__pb2
|
|
from google.protobuf import struct_pb2 as google_dot_protobuf_dot_struct__pb2
|
|
|
|
|
|
DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(b'\n\x16sglang_scheduler.proto\x12\x15sglang.grpc.scheduler\x1a\x1fgoogle/protobuf/timestamp.proto\x1a\x1cgoogle/protobuf/struct.proto\"\xe1\x05\n\x0eSamplingParams\x12\x13\n\x0btemperature\x18\x01 \x01(\x02\x12\r\n\x05top_p\x18\x02 \x01(\x02\x12\r\n\x05top_k\x18\x03 \x01(\x05\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\x1b\n\x0emax_new_tokens\x18\x08 \x01(\x05H\x01\x88\x01\x01\x12\x0c\n\x04stop\x18\t \x03(\t\x12\x16\n\x0estop_token_ids\x18\n \x03(\r\x12\x1b\n\x13skip_special_tokens\x18\x0b \x01(\x08\x12%\n\x1dspaces_between_special_tokens\x18\x0c \x01(\x08\x12\x0f\n\x05regex\x18\r \x01(\tH\x00\x12\x15\n\x0bjson_schema\x18\x0e \x01(\tH\x00\x12\x16\n\x0c\x65\x62nf_grammar\x18\x0f \x01(\tH\x00\x12\x18\n\x0estructural_tag\x18\x10 \x01(\tH\x00\x12\x11\n\tlora_path\x18\x11 \x01(\t\x12\t\n\x01n\x18\x12 \x01(\x05\x12\x15\n\rtoken_healing\x18\x13 \x01(\x08\x12\x16\n\x0emin_new_tokens\x18\x14 \x01(\x05\x12\x12\n\nignore_eos\x18\x15 \x01(\x08\x12\x14\n\x0cno_stop_trim\x18\x16 \x01(\x08\x12\x17\n\x0fstream_interval\x18\x17 \x01(\x05\x12H\n\nlogit_bias\x18\x18 \x03(\x0b\x32\x34.sglang.grpc.scheduler.SamplingParams.LogitBiasEntry\x12.\n\rcustom_params\x18\x19 \x01(\x0b\x32\x17.google.protobuf.Struct\x1a\x30\n\x0eLogitBiasEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\r\n\x05value\x18\x02 \x01(\x02:\x02\x38\x01\x42\x0c\n\nconstraintB\x11\n\x0f_max_new_tokens\"]\n\x13\x44isaggregatedParams\x12\x16\n\x0e\x62ootstrap_host\x18\x01 \x01(\t\x12\x16\n\x0e\x62ootstrap_port\x18\x02 \x01(\x05\x12\x16\n\x0e\x62ootstrap_room\x18\x03 \x01(\x05\"\xe2\x04\n\x0fGenerateRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\ttokenized\x18\x02 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\x12:\n\tmm_inputs\x18\x03 \x01(\x0b\x32\'.sglang.grpc.scheduler.MultimodalInputs\x12>\n\x0fsampling_params\x18\x04 \x01(\x0b\x32%.sglang.grpc.scheduler.SamplingParams\x12\x16\n\x0ereturn_logprob\x18\x05 \x01(\x08\x12\x19\n\x11logprob_start_len\x18\x06 \x01(\x05\x12\x18\n\x10top_logprobs_num\x18\x07 \x01(\x05\x12\x19\n\x11token_ids_logprob\x18\x08 \x03(\r\x12\x1c\n\x14return_hidden_states\x18\t \x01(\x08\x12H\n\x14\x64isaggregated_params\x18\n \x01(\x0b\x32*.sglang.grpc.scheduler.DisaggregatedParams\x12\x1e\n\x16\x63ustom_logit_processor\x18\x0b \x01(\t\x12-\n\ttimestamp\x18\x0c \x01(\x0b\x32\x1a.google.protobuf.Timestamp\x12\x13\n\x0blog_metrics\x18\r \x01(\x08\x12\x14\n\x0cinput_embeds\x18\x0e \x03(\x02\x12\x0f\n\x07lora_id\x18\x0f \x01(\t\x12\x1a\n\x12\x64\x61ta_parallel_rank\x18\x10 \x01(\x05\x12\x0e\n\x06stream\x18\x11 \x01(\x08\":\n\x0eTokenizedInput\x12\x15\n\roriginal_text\x18\x01 \x01(\t\x12\x11\n\tinput_ids\x18\x02 \x03(\r\"\xd3\x01\n\x10MultimodalInputs\x12\x12\n\nimage_urls\x18\x01 \x03(\t\x12\x12\n\nvideo_urls\x18\x02 \x03(\t\x12\x12\n\naudio_urls\x18\x03 \x03(\t\x12\x33\n\x12processed_features\x18\x04 \x01(\x0b\x32\x17.google.protobuf.Struct\x12\x12\n\nimage_data\x18\x05 \x03(\x0c\x12\x12\n\nvideo_data\x18\x06 \x03(\x0c\x12\x12\n\naudio_data\x18\x07 \x03(\x0c\x12\x12\n\nmodalities\x18\x08 \x03(\t\"\xe3\x01\n\x10GenerateResponse\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12;\n\x05\x63hunk\x18\x02 \x01(\x0b\x32*.sglang.grpc.scheduler.GenerateStreamChunkH\x00\x12;\n\x08\x63omplete\x18\x03 \x01(\x0b\x32\'.sglang.grpc.scheduler.GenerateCompleteH\x00\x12\x35\n\x05\x65rror\x18\x04 \x01(\x0b\x32$.sglang.grpc.scheduler.GenerateErrorH\x00\x42\n\n\x08response\"\x86\x02\n\x13GenerateStreamChunk\x12\x11\n\ttoken_ids\x18\x01 \x03(\r\x12\x15\n\rprompt_tokens\x18\x02 \x01(\x05\x12\x19\n\x11\x63ompletion_tokens\x18\x03 \x01(\x05\x12\x15\n\rcached_tokens\x18\x04 \x01(\x05\x12>\n\x0foutput_logprobs\x18\x05 \x01(\x0b\x32%.sglang.grpc.scheduler.OutputLogProbs\x12\x15\n\rhidden_states\x18\x06 \x03(\x02\x12<\n\x0einput_logprobs\x18\x07 \x01(\x0b\x32$.sglang.grpc.scheduler.InputLogProbs\"\x8c\x03\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(\x05\x12\x19\n\x11\x63ompletion_tokens\x18\x04 \x01(\x05\x12\x15\n\rcached_tokens\x18\x05 \x01(\x05\x12>\n\x0foutput_logprobs\x18\x06 \x01(\x0b\x32%.sglang.grpc.scheduler.OutputLogProbs\x12>\n\x11\x61ll_hidden_states\x18\x07 \x03(\x0b\x32#.sglang.grpc.scheduler.HiddenStates\x12\x1a\n\x10matched_token_id\x18\x08 \x01(\rH\x00\x12\x1a\n\x10matched_stop_str\x18\t \x01(\tH\x00\x12<\n\x0einput_logprobs\x18\n \x01(\x0b\x32$.sglang.grpc.scheduler.InputLogProbsB\x0e\n\x0cmatched_stop\"K\n\rGenerateError\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x18\n\x10http_status_code\x18\x02 \x01(\t\x12\x0f\n\x07\x64\x65tails\x18\x03 \x01(\t\"u\n\x0eOutputLogProbs\x12\x16\n\x0etoken_logprobs\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x38\n\x0ctop_logprobs\x18\x03 \x03(\x0b\x32\".sglang.grpc.scheduler.TopLogProbs\"\x9e\x01\n\rInputLogProbs\x12@\n\x0etoken_logprobs\x18\x01 \x03(\x0b\x32(.sglang.grpc.scheduler.InputTokenLogProb\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\x12\x38\n\x0ctop_logprobs\x18\x03 \x03(\x0b\x32\".sglang.grpc.scheduler.TopLogProbs\"1\n\x11InputTokenLogProb\x12\x12\n\x05value\x18\x01 \x01(\x02H\x00\x88\x01\x01\x42\x08\n\x06_value\"0\n\x0bTopLogProbs\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\x11\n\ttoken_ids\x18\x02 \x03(\x05\"?\n\x0cHiddenStates\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\r\n\x05layer\x18\x02 \x01(\x05\x12\x10\n\x08position\x18\x03 \x01(\x05\"\xca\x02\n\x0c\x45mbedRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\ttokenized\x18\x02 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\x12:\n\tmm_inputs\x18\x04 \x01(\x0b\x32\'.sglang.grpc.scheduler.MultimodalInputs\x12>\n\x0fsampling_params\x18\x05 \x01(\x0b\x32%.sglang.grpc.scheduler.SamplingParams\x12\x13\n\x0blog_metrics\x18\x06 \x01(\x08\x12\x16\n\x0etoken_type_ids\x18\x07 \x03(\x05\x12\x1a\n\x12\x64\x61ta_parallel_rank\x18\x08 \x01(\x05\x12\x18\n\x10is_cross_encoder\x18\t \x01(\x08\x12\r\n\x05texts\x18\n \x03(\t\"\x9d\x01\n\rEmbedResponse\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x38\n\x08\x63omplete\x18\x02 \x01(\x0b\x32$.sglang.grpc.scheduler.EmbedCompleteH\x00\x12\x32\n\x05\x65rror\x18\x03 \x01(\x0b\x32!.sglang.grpc.scheduler.EmbedErrorH\x00\x42\n\n\x08response\"\xa3\x01\n\rEmbedComplete\x12\x11\n\tembedding\x18\x01 \x03(\x02\x12\x15\n\rprompt_tokens\x18\x02 \x01(\x05\x12\x15\n\rcached_tokens\x18\x03 \x01(\x05\x12\x15\n\rembedding_dim\x18\x04 \x01(\x05\x12:\n\x10\x62\x61tch_embeddings\x18\x05 \x03(\x0b\x32 .sglang.grpc.scheduler.Embedding\"*\n\tEmbedding\x12\x0e\n\x06values\x18\x01 \x03(\x02\x12\r\n\x05index\x18\x02 \x01(\x05\"<\n\nEmbedError\x12\x0f\n\x07message\x18\x01 \x01(\t\x12\x0c\n\x04\x63ode\x18\x02 \x01(\t\x12\x0f\n\x07\x64\x65tails\x18\x03 \x01(\t\"N\n\x12HealthCheckRequest\x12\x38\n\ttokenized\x18\x01 \x01(\x0b\x32%.sglang.grpc.scheduler.TokenizedInput\"7\n\x13HealthCheckResponse\x12\x0f\n\x07healthy\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"2\n\x0c\x41\x62ortRequest\x12\x12\n\nrequest_id\x18\x01 \x01(\t\x12\x0e\n\x06reason\x18\x02 \x01(\t\"1\n\rAbortResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"I\n\x0fLoadLoRARequest\x12\x12\n\nadapter_id\x18\x01 \x01(\t\x12\x14\n\x0c\x61\x64\x61pter_path\x18\x02 \x01(\t\x12\x0c\n\x04rank\x18\x03 \x01(\x05\"H\n\x10LoadLoRAResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x12\n\nadapter_id\x18\x02 \x01(\t\x12\x0f\n\x07message\x18\x03 \x01(\t\"\'\n\x11UnloadLoRARequest\x12\x12\n\nadapter_id\x18\x01 \x01(\t\"6\n\x12UnloadLoRAResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"w\n\x14UpdateWeightsRequest\x12\x13\n\tdisk_path\x18\x01 \x01(\tH\x00\x12\x15\n\x0btensor_data\x18\x02 \x01(\x0cH\x00\x12\x14\n\nremote_url\x18\x03 \x01(\tH\x00\x12\x13\n\x0bweight_name\x18\x04 \x01(\tB\x08\n\x06source\"9\n\x15UpdateWeightsResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t\"-\n\x17GetInternalStateRequest\x12\x12\n\nstate_keys\x18\x01 \x03(\t\"B\n\x18GetInternalStateResponse\x12&\n\x05state\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\"A\n\x17SetInternalStateRequest\x12&\n\x05state\x18\x01 \x01(\x0b\x32\x17.google.protobuf.Struct\"<\n\x18SetInternalStateResponse\x12\x0f\n\x07success\x18\x01 \x01(\x08\x12\x0f\n\x07message\x18\x02 \x01(\t2\xfe\x02\n\x0fSglangScheduler\x12]\n\x08Generate\x12&.sglang.grpc.scheduler.GenerateRequest\x1a\'.sglang.grpc.scheduler.GenerateResponse0\x01\x12R\n\x05\x45mbed\x12#.sglang.grpc.scheduler.EmbedRequest\x1a$.sglang.grpc.scheduler.EmbedResponse\x12\x64\n\x0bHealthCheck\x12).sglang.grpc.scheduler.HealthCheckRequest\x1a*.sglang.grpc.scheduler.HealthCheckResponse\x12R\n\x05\x41\x62ort\x12#.sglang.grpc.scheduler.AbortRequest\x1a$.sglang.grpc.scheduler.AbortResponseb\x06proto3')
|
|
|
|
_globals = globals()
|
|
_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
|
|
_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, 'sglang_scheduler_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=113
|
|
_globals['_SAMPLINGPARAMS']._serialized_end=850
|
|
_globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_start=769
|
|
_globals['_SAMPLINGPARAMS_LOGITBIASENTRY']._serialized_end=817
|
|
_globals['_DISAGGREGATEDPARAMS']._serialized_start=852
|
|
_globals['_DISAGGREGATEDPARAMS']._serialized_end=945
|
|
_globals['_GENERATEREQUEST']._serialized_start=948
|
|
_globals['_GENERATEREQUEST']._serialized_end=1558
|
|
_globals['_TOKENIZEDINPUT']._serialized_start=1560
|
|
_globals['_TOKENIZEDINPUT']._serialized_end=1618
|
|
_globals['_MULTIMODALINPUTS']._serialized_start=1621
|
|
_globals['_MULTIMODALINPUTS']._serialized_end=1832
|
|
_globals['_GENERATERESPONSE']._serialized_start=1835
|
|
_globals['_GENERATERESPONSE']._serialized_end=2062
|
|
_globals['_GENERATESTREAMCHUNK']._serialized_start=2065
|
|
_globals['_GENERATESTREAMCHUNK']._serialized_end=2327
|
|
_globals['_GENERATECOMPLETE']._serialized_start=2330
|
|
_globals['_GENERATECOMPLETE']._serialized_end=2726
|
|
_globals['_GENERATEERROR']._serialized_start=2728
|
|
_globals['_GENERATEERROR']._serialized_end=2803
|
|
_globals['_OUTPUTLOGPROBS']._serialized_start=2805
|
|
_globals['_OUTPUTLOGPROBS']._serialized_end=2922
|
|
_globals['_INPUTLOGPROBS']._serialized_start=2925
|
|
_globals['_INPUTLOGPROBS']._serialized_end=3083
|
|
_globals['_INPUTTOKENLOGPROB']._serialized_start=3085
|
|
_globals['_INPUTTOKENLOGPROB']._serialized_end=3134
|
|
_globals['_TOPLOGPROBS']._serialized_start=3136
|
|
_globals['_TOPLOGPROBS']._serialized_end=3184
|
|
_globals['_HIDDENSTATES']._serialized_start=3186
|
|
_globals['_HIDDENSTATES']._serialized_end=3249
|
|
_globals['_EMBEDREQUEST']._serialized_start=3252
|
|
_globals['_EMBEDREQUEST']._serialized_end=3582
|
|
_globals['_EMBEDRESPONSE']._serialized_start=3585
|
|
_globals['_EMBEDRESPONSE']._serialized_end=3742
|
|
_globals['_EMBEDCOMPLETE']._serialized_start=3745
|
|
_globals['_EMBEDCOMPLETE']._serialized_end=3908
|
|
_globals['_EMBEDDING']._serialized_start=3910
|
|
_globals['_EMBEDDING']._serialized_end=3952
|
|
_globals['_EMBEDERROR']._serialized_start=3954
|
|
_globals['_EMBEDERROR']._serialized_end=4014
|
|
_globals['_HEALTHCHECKREQUEST']._serialized_start=4016
|
|
_globals['_HEALTHCHECKREQUEST']._serialized_end=4094
|
|
_globals['_HEALTHCHECKRESPONSE']._serialized_start=4096
|
|
_globals['_HEALTHCHECKRESPONSE']._serialized_end=4151
|
|
_globals['_ABORTREQUEST']._serialized_start=4153
|
|
_globals['_ABORTREQUEST']._serialized_end=4203
|
|
_globals['_ABORTRESPONSE']._serialized_start=4205
|
|
_globals['_ABORTRESPONSE']._serialized_end=4254
|
|
_globals['_LOADLORAREQUEST']._serialized_start=4256
|
|
_globals['_LOADLORAREQUEST']._serialized_end=4329
|
|
_globals['_LOADLORARESPONSE']._serialized_start=4331
|
|
_globals['_LOADLORARESPONSE']._serialized_end=4403
|
|
_globals['_UNLOADLORAREQUEST']._serialized_start=4405
|
|
_globals['_UNLOADLORAREQUEST']._serialized_end=4444
|
|
_globals['_UNLOADLORARESPONSE']._serialized_start=4446
|
|
_globals['_UNLOADLORARESPONSE']._serialized_end=4500
|
|
_globals['_UPDATEWEIGHTSREQUEST']._serialized_start=4502
|
|
_globals['_UPDATEWEIGHTSREQUEST']._serialized_end=4621
|
|
_globals['_UPDATEWEIGHTSRESPONSE']._serialized_start=4623
|
|
_globals['_UPDATEWEIGHTSRESPONSE']._serialized_end=4680
|
|
_globals['_GETINTERNALSTATEREQUEST']._serialized_start=4682
|
|
_globals['_GETINTERNALSTATEREQUEST']._serialized_end=4727
|
|
_globals['_GETINTERNALSTATERESPONSE']._serialized_start=4729
|
|
_globals['_GETINTERNALSTATERESPONSE']._serialized_end=4795
|
|
_globals['_SETINTERNALSTATEREQUEST']._serialized_start=4797
|
|
_globals['_SETINTERNALSTATEREQUEST']._serialized_end=4862
|
|
_globals['_SETINTERNALSTATERESPONSE']._serialized_start=4864
|
|
_globals['_SETINTERNALSTATERESPONSE']._serialized_end=4924
|
|
_globals['_SGLANGSCHEDULER']._serialized_start=4927
|
|
_globals['_SGLANGSCHEDULER']._serialized_end=5309
|
|
# @@protoc_insertion_point(module_scope)
|