vllm-cpu-avx512bf16 0.14.0__cp313-cp313-manylinux_2_28_x86_64.whl

vllm/grpc/vllm_engine.proto

@@ -0,0 +1,195 @@
+syntax = "proto3";
+
+package vllm.grpc.engine;
+
+// Service definition for vLLM engine communication
+// This protocol is designed for efficient binary communication between
+// the Rust router and vLLM Python engine (AsyncLLM).
+service VllmEngine {
+  // Submit a generation request (supports streaming)
+  rpc Generate(GenerateRequest) returns (stream GenerateResponse);
+
+  // Submit an embedding request
+  rpc Embed(EmbedRequest) returns (EmbedResponse);
+
+  // Health check
+  rpc HealthCheck(HealthCheckRequest) returns (HealthCheckResponse);
+
+  // Abort a running request
+  rpc Abort(AbortRequest) returns (AbortResponse);
+
+  // Get model information
+  rpc GetModelInfo(GetModelInfoRequest) returns (GetModelInfoResponse);
+
+  // Get server information
+  rpc GetServerInfo(GetServerInfoRequest) returns (GetServerInfoResponse);
+}
+
+// =====================
+// Common Types
+// =====================
+
+// Sampling parameters for text generation
+message SamplingParams {
+  optional float temperature = 1;
+  float top_p = 2;
+  uint32 top_k = 3;
+  float min_p = 4;
+  float frequency_penalty = 5;
+  float presence_penalty = 6;
+  float repetition_penalty = 7;
+
+  optional uint32 max_tokens = 8;
+  uint32 min_tokens = 9;
+
+  repeated string stop = 10;
+  repeated uint32 stop_token_ids = 11;
+
+  bool skip_special_tokens = 12;
+  bool spaces_between_special_tokens = 13;
+  bool ignore_eos = 14;
+
+  uint32 n = 15;  // Number of parallel samples
+
+  // Logprobs configuration
+  optional int32 logprobs = 22;  // Number of log probabilities per output token (-1 for all)
+  optional int32 prompt_logprobs = 23;  // Number of log probabilities per prompt token (-1 for all)
+
+  // Additional vLLM fields
+  optional int32 seed = 24;  // Random seed for reproducibility
+  bool include_stop_str_in_output = 25;  // Whether to include stop strings in output
+  map<int32, float> logit_bias = 26;  // Token ID to bias mapping (-100 to 100)
+  optional int32 truncate_prompt_tokens = 27;  // Prompt truncation (-1 for model max)
+
+  // Structured outputs (one of) - matches vLLM's StructuredOutputsParams
+  oneof constraint {
+    string json_schema = 16;  // JSON schema for structured output
+    string regex = 17;  // Regex pattern
+    string grammar = 18;  // Grammar/EBNF for structured output
+    string structural_tag = 19;  // Structural tag (e.g., Harmony models)
+    bool json_object = 20;  // Force JSON object output
+    ChoiceConstraint choice = 21;  // List of allowed choices
+  }
+}
+
+// Choice constraint for structured outputs
+message ChoiceConstraint {
+  repeated string choices = 1;
+}
+
+// Pre-tokenized input from Rust router
+message TokenizedInput {
+  string original_text = 1;  // For reference/debugging
+  repeated uint32 input_ids = 2;  // Actual token IDs to process
+}
+
+// =====================
+// Generate Request
+// =====================
+
+message GenerateRequest {
+  string request_id = 1;
+
+  // Prompt input
+  oneof input {
+    TokenizedInput tokenized = 2;
+    string text = 3;
+  }
+
+  // Generation parameters (includes logprobs config)
+  SamplingParams sampling_params = 4;
+
+  // Streaming
+  bool stream = 5;
+}
+
+// =====================
+// Generate Response
+// =====================
+
+message GenerateResponse {
+  oneof response {
+    GenerateStreamChunk chunk = 1;     // For streaming
+    GenerateComplete complete = 2;     // For final/non-streaming
+  }
+}
+
+message GenerateStreamChunk {
+  repeated uint32 token_ids = 1;       // Incremental tokens
+  uint32 prompt_tokens = 2;
+  uint32 completion_tokens = 3;
+  uint32 cached_tokens = 4;
+
+  // Logprobs support (TODO: implement in Phase 4)
+  // OutputLogProbs output_logprobs = 5;
+  // InputLogProbs input_logprobs = 6;  // Only in first chunk
+}
+
+message GenerateComplete {
+  repeated uint32 output_ids = 1;      // All output tokens
+  string finish_reason = 2;            // "stop", "length", "abort"
+  uint32 prompt_tokens = 3;
+  uint32 completion_tokens = 4;
+  uint32 cached_tokens = 5;
+
+  // Logprobs support (TODO: implement in Phase 4)
+  // OutputLogProbs output_logprobs = 6;
+  // InputLogProbs input_logprobs = 7;
+}
+
+// =====================
+// Embedding Request
+// =====================
+
+message EmbedRequest {
+  string request_id = 1;
+  TokenizedInput tokenized = 2;
+}
+
+message EmbedResponse {
+  repeated float embedding = 1;
+  uint32 prompt_tokens = 2;
+  uint32 embedding_dim = 3;
+}
+
+// =====================
+// Management Operations
+// =====================
+
+message HealthCheckRequest {}
+
+message HealthCheckResponse {
+  bool healthy = 1;
+  string message = 2;
+}
+
+message AbortRequest {
+  repeated string request_ids = 1;
+}
+
+message AbortResponse {
+}
+
+// =====================
+// Model and Server Info
+// =====================
+
+message GetModelInfoRequest {}
+
+message GetModelInfoResponse {
+  string model_path = 1;
+  bool is_generation = 2;
+  uint32 max_context_length = 3;
+  uint32 vocab_size = 4;
+  bool supports_vision = 5;
+}
+
+message GetServerInfoRequest {}
+
+message GetServerInfoResponse {
+  uint32 active_requests = 1;
+  bool is_paused = 2;
+  double last_receive_timestamp = 3;
+  double uptime_seconds = 4;
+  string server_type = 5;  // "vllm-grpc"
+}

vllm/grpc/vllm_engine_pb2.py

@@ -0,0 +1,77 @@
+# 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)

vllm/grpc/vllm_engine_pb2.pyi

@@ -0,0 +1,213 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+# mypy: ignore-errors
+from google.protobuf.internal import containers as _containers
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import message as _message
+from collections.abc import Iterable as _Iterable, Mapping as _Mapping
+from typing import ClassVar as _ClassVar, Optional as _Optional, Union as _Union
+
+DESCRIPTOR: _descriptor.FileDescriptor
+
+class SamplingParams(_message.Message):
+    __slots__ = ("temperature", "top_p", "top_k", "min_p", "frequency_penalty", "presence_penalty", "repetition_penalty", "max_tokens", "min_tokens", "stop", "stop_token_ids", "skip_special_tokens", "spaces_between_special_tokens", "ignore_eos", "n", "logprobs", "prompt_logprobs", "seed", "include_stop_str_in_output", "logit_bias", "truncate_prompt_tokens", "json_schema", "regex", "grammar", "structural_tag", "json_object", "choice")
+    class LogitBiasEntry(_message.Message):
+        __slots__ = ("key", "value")
+        KEY_FIELD_NUMBER: _ClassVar[int]
+        VALUE_FIELD_NUMBER: _ClassVar[int]
+        key: int
+        value: float
+        def __init__(self, key: _Optional[int] = ..., value: _Optional[float] = ...) -> None: ...
+    TEMPERATURE_FIELD_NUMBER: _ClassVar[int]
+    TOP_P_FIELD_NUMBER: _ClassVar[int]
+    TOP_K_FIELD_NUMBER: _ClassVar[int]
+    MIN_P_FIELD_NUMBER: _ClassVar[int]
+    FREQUENCY_PENALTY_FIELD_NUMBER: _ClassVar[int]
+    PRESENCE_PENALTY_FIELD_NUMBER: _ClassVar[int]
+    REPETITION_PENALTY_FIELD_NUMBER: _ClassVar[int]
+    MAX_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    MIN_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    STOP_FIELD_NUMBER: _ClassVar[int]
+    STOP_TOKEN_IDS_FIELD_NUMBER: _ClassVar[int]
+    SKIP_SPECIAL_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    SPACES_BETWEEN_SPECIAL_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    IGNORE_EOS_FIELD_NUMBER: _ClassVar[int]
+    N_FIELD_NUMBER: _ClassVar[int]
+    LOGPROBS_FIELD_NUMBER: _ClassVar[int]
+    PROMPT_LOGPROBS_FIELD_NUMBER: _ClassVar[int]
+    SEED_FIELD_NUMBER: _ClassVar[int]
+    INCLUDE_STOP_STR_IN_OUTPUT_FIELD_NUMBER: _ClassVar[int]
+    LOGIT_BIAS_FIELD_NUMBER: _ClassVar[int]
+    TRUNCATE_PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    JSON_SCHEMA_FIELD_NUMBER: _ClassVar[int]
+    REGEX_FIELD_NUMBER: _ClassVar[int]
+    GRAMMAR_FIELD_NUMBER: _ClassVar[int]
+    STRUCTURAL_TAG_FIELD_NUMBER: _ClassVar[int]
+    JSON_OBJECT_FIELD_NUMBER: _ClassVar[int]
+    CHOICE_FIELD_NUMBER: _ClassVar[int]
+    temperature: float
+    top_p: float
+    top_k: int
+    min_p: float
+    frequency_penalty: float
+    presence_penalty: float
+    repetition_penalty: float
+    max_tokens: int
+    min_tokens: int
+    stop: _containers.RepeatedScalarFieldContainer[str]
+    stop_token_ids: _containers.RepeatedScalarFieldContainer[int]
+    skip_special_tokens: bool
+    spaces_between_special_tokens: bool
+    ignore_eos: bool
+    n: int
+    logprobs: int
+    prompt_logprobs: int
+    seed: int
+    include_stop_str_in_output: bool
+    logit_bias: _containers.ScalarMap[int, float]
+    truncate_prompt_tokens: int
+    json_schema: str
+    regex: str
+    grammar: str
+    structural_tag: str
+    json_object: bool
+    choice: ChoiceConstraint
+    def __init__(self, temperature: _Optional[float] = ..., top_p: _Optional[float] = ..., top_k: _Optional[int] = ..., min_p: _Optional[float] = ..., frequency_penalty: _Optional[float] = ..., presence_penalty: _Optional[float] = ..., repetition_penalty: _Optional[float] = ..., max_tokens: _Optional[int] = ..., min_tokens: _Optional[int] = ..., stop: _Optional[_Iterable[str]] = ..., stop_token_ids: _Optional[_Iterable[int]] = ..., skip_special_tokens: bool = ..., spaces_between_special_tokens: bool = ..., ignore_eos: bool = ..., n: _Optional[int] = ..., logprobs: _Optional[int] = ..., prompt_logprobs: _Optional[int] = ..., seed: _Optional[int] = ..., include_stop_str_in_output: bool = ..., logit_bias: _Optional[_Mapping[int, float]] = ..., truncate_prompt_tokens: _Optional[int] = ..., json_schema: _Optional[str] = ..., regex: _Optional[str] = ..., grammar: _Optional[str] = ..., structural_tag: _Optional[str] = ..., json_object: bool = ..., choice: _Optional[_Union[ChoiceConstraint, _Mapping]] = ...) -> None: ...
+
+class ChoiceConstraint(_message.Message):
+    __slots__ = ("choices",)
+    CHOICES_FIELD_NUMBER: _ClassVar[int]
+    choices: _containers.RepeatedScalarFieldContainer[str]
+    def __init__(self, choices: _Optional[_Iterable[str]] = ...) -> None: ...
+
+class TokenizedInput(_message.Message):
+    __slots__ = ("original_text", "input_ids")
+    ORIGINAL_TEXT_FIELD_NUMBER: _ClassVar[int]
+    INPUT_IDS_FIELD_NUMBER: _ClassVar[int]
+    original_text: str
+    input_ids: _containers.RepeatedScalarFieldContainer[int]
+    def __init__(self, original_text: _Optional[str] = ..., input_ids: _Optional[_Iterable[int]] = ...) -> None: ...
+
+class GenerateRequest(_message.Message):
+    __slots__ = ("request_id", "tokenized", "text", "sampling_params", "stream")
+    REQUEST_ID_FIELD_NUMBER: _ClassVar[int]
+    TOKENIZED_FIELD_NUMBER: _ClassVar[int]
+    TEXT_FIELD_NUMBER: _ClassVar[int]
+    SAMPLING_PARAMS_FIELD_NUMBER: _ClassVar[int]
+    STREAM_FIELD_NUMBER: _ClassVar[int]
+    request_id: str
+    tokenized: TokenizedInput
+    text: str
+    sampling_params: SamplingParams
+    stream: bool
+    def __init__(self, request_id: _Optional[str] = ..., tokenized: _Optional[_Union[TokenizedInput, _Mapping]] = ..., text: _Optional[str] = ..., sampling_params: _Optional[_Union[SamplingParams, _Mapping]] = ..., stream: bool = ...) -> None: ...
+
+class GenerateResponse(_message.Message):
+    __slots__ = ("chunk", "complete")
+    CHUNK_FIELD_NUMBER: _ClassVar[int]
+    COMPLETE_FIELD_NUMBER: _ClassVar[int]
+    chunk: GenerateStreamChunk
+    complete: GenerateComplete
+    def __init__(self, chunk: _Optional[_Union[GenerateStreamChunk, _Mapping]] = ..., complete: _Optional[_Union[GenerateComplete, _Mapping]] = ...) -> None: ...
+
+class GenerateStreamChunk(_message.Message):
+    __slots__ = ("token_ids", "prompt_tokens", "completion_tokens", "cached_tokens")
+    TOKEN_IDS_FIELD_NUMBER: _ClassVar[int]
+    PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    COMPLETION_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    CACHED_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    token_ids: _containers.RepeatedScalarFieldContainer[int]
+    prompt_tokens: int
+    completion_tokens: int
+    cached_tokens: int
+    def __init__(self, token_ids: _Optional[_Iterable[int]] = ..., prompt_tokens: _Optional[int] = ..., completion_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ...) -> None: ...
+
+class GenerateComplete(_message.Message):
+    __slots__ = ("output_ids", "finish_reason", "prompt_tokens", "completion_tokens", "cached_tokens")
+    OUTPUT_IDS_FIELD_NUMBER: _ClassVar[int]
+    FINISH_REASON_FIELD_NUMBER: _ClassVar[int]
+    PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    COMPLETION_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    CACHED_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    output_ids: _containers.RepeatedScalarFieldContainer[int]
+    finish_reason: str
+    prompt_tokens: int
+    completion_tokens: int
+    cached_tokens: int
+    def __init__(self, output_ids: _Optional[_Iterable[int]] = ..., finish_reason: _Optional[str] = ..., prompt_tokens: _Optional[int] = ..., completion_tokens: _Optional[int] = ..., cached_tokens: _Optional[int] = ...) -> None: ...
+
+class EmbedRequest(_message.Message):
+    __slots__ = ("request_id", "tokenized")
+    REQUEST_ID_FIELD_NUMBER: _ClassVar[int]
+    TOKENIZED_FIELD_NUMBER: _ClassVar[int]
+    request_id: str
+    tokenized: TokenizedInput
+    def __init__(self, request_id: _Optional[str] = ..., tokenized: _Optional[_Union[TokenizedInput, _Mapping]] = ...) -> None: ...
+
+class EmbedResponse(_message.Message):
+    __slots__ = ("embedding", "prompt_tokens", "embedding_dim")
+    EMBEDDING_FIELD_NUMBER: _ClassVar[int]
+    PROMPT_TOKENS_FIELD_NUMBER: _ClassVar[int]
+    EMBEDDING_DIM_FIELD_NUMBER: _ClassVar[int]
+    embedding: _containers.RepeatedScalarFieldContainer[float]
+    prompt_tokens: int
+    embedding_dim: int
+    def __init__(self, embedding: _Optional[_Iterable[float]] = ..., prompt_tokens: _Optional[int] = ..., embedding_dim: _Optional[int] = ...) -> None: ...
+
+class HealthCheckRequest(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class HealthCheckResponse(_message.Message):
+    __slots__ = ("healthy", "message")
+    HEALTHY_FIELD_NUMBER: _ClassVar[int]
+    MESSAGE_FIELD_NUMBER: _ClassVar[int]
+    healthy: bool
+    message: str
+    def __init__(self, healthy: bool = ..., message: _Optional[str] = ...) -> None: ...
+
+class AbortRequest(_message.Message):
+    __slots__ = ("request_ids",)
+    REQUEST_IDS_FIELD_NUMBER: _ClassVar[int]
+    request_ids: _containers.RepeatedScalarFieldContainer[str]
+    def __init__(self, request_ids: _Optional[_Iterable[str]] = ...) -> None: ...
+
+class AbortResponse(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class GetModelInfoRequest(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class GetModelInfoResponse(_message.Message):
+    __slots__ = ("model_path", "is_generation", "max_context_length", "vocab_size", "supports_vision")
+    MODEL_PATH_FIELD_NUMBER: _ClassVar[int]
+    IS_GENERATION_FIELD_NUMBER: _ClassVar[int]
+    MAX_CONTEXT_LENGTH_FIELD_NUMBER: _ClassVar[int]
+    VOCAB_SIZE_FIELD_NUMBER: _ClassVar[int]
+    SUPPORTS_VISION_FIELD_NUMBER: _ClassVar[int]
+    model_path: str
+    is_generation: bool
+    max_context_length: int
+    vocab_size: int
+    supports_vision: bool
+    def __init__(self, model_path: _Optional[str] = ..., is_generation: bool = ..., max_context_length: _Optional[int] = ..., vocab_size: _Optional[int] = ..., supports_vision: bool = ...) -> None: ...
+
+class GetServerInfoRequest(_message.Message):
+    __slots__ = ()
+    def __init__(self) -> None: ...
+
+class GetServerInfoResponse(_message.Message):
+    __slots__ = ("active_requests", "is_paused", "last_receive_timestamp", "uptime_seconds", "server_type")
+    ACTIVE_REQUESTS_FIELD_NUMBER: _ClassVar[int]
+    IS_PAUSED_FIELD_NUMBER: _ClassVar[int]
+    LAST_RECEIVE_TIMESTAMP_FIELD_NUMBER: _ClassVar[int]
+    UPTIME_SECONDS_FIELD_NUMBER: _ClassVar[int]
+    SERVER_TYPE_FIELD_NUMBER: _ClassVar[int]
+    active_requests: int
+    is_paused: bool
+    last_receive_timestamp: float
+    uptime_seconds: float
+    server_type: str
+    def __init__(self, active_requests: _Optional[int] = ..., is_paused: bool = ..., last_receive_timestamp: _Optional[float] = ..., uptime_seconds: _Optional[float] = ..., server_type: _Optional[str] = ...) -> None: ...