server.proto

syntax = "proto3";

package llm_perf;

// Containers to hold repeated fundamental values.
message FloatList {
    repeated double values = 1;
}

message Int64List {
    repeated int64 values = 1;
}

message BytesList {
    repeated bytes values = 1;
}

message StringList {
    repeated string values = 1;
}

message Struct {
  // Unordered map of dynamically typed values.
  map<string, Value> fields = 1;
}

// Container for non-sequential data.
message Value {
  oneof kind {
    FloatList float_list = 1;
    Int64List int64_list = 2;
    BytesList bytes_list = 3;
    float float_ = 4;
    int64 int64_ = 5;
    bytes bytes_ = 6;
    StringList string_list = 7;
    string string_ = 8;
    Struct struct_ = 9;
  }
}

message InferenceRequest {
  string req_id = 1;
  map<string, Value> inputs = 2;
}

message InferenceResponse {
  string req_id = 1;
  map<string, Value> outputs = 2;
}

service Inference {
  // 1 request -> 1 response
  rpc CompleteInference(InferenceRequest) returns (InferenceResponse) {}
  // 1 request -> m response
  rpc StreamingInference(InferenceRequest) returns (stream InferenceResponse) {}
}