Chat Proto

chat.proto

syntax = "proto3";

package chat;

option go_package = "github.com/defenseunicorns/leapfrogai/pkg/client/chat";

// ChatRequest is the payload to Chat creation
message ChatRequest {
    repeated string inputs = 1;
}

message Chat {
    repeated float chat = 1;
}

// ChatResponse are what's returned by the gRPC service
message ChatResponse {
    repeated Chat chat = 1;
}

service ChatsService {
    rpc ChatCompletion(ChatRequest) returns (ChatResponse);
    rpc ChatCompletionStream(ChatRequest) returns (stream ChatResponse);
}

// Generic OpenAI
enum OpenAIChatRole {
    USER = 0;
    SYSTEM = 1;
    FUNCTION = 2;
    ASSISTANT = 3;
}

message OpenAIChatItem {
  OpenAIChatRole role = 1;
  string content = 2;
}

// OpenAI ChatCompletion Request
message OpenAIChatCompletionRequest {
    string model = 1; // added
    repeated OpenAIChatItem chat_items = 2; // added
    float temperature = 3; // added
    float top_p = 4; // added
    int32 n = 5; //added
    bool stream = 7; // will be used to route to either 
    repeated string stop = 8; //added
    int32 max_tokens = 9; // max_new_tokens
    float presence_penalty = 10;  // added
    float frequency_penalty = 11; // added
    map<string, int32> logit_bias = 12; // added
    string user = 13; // added
}

// OpenAI ChatCompletion Request
message OpenAICompletionChoice {
    int32 index = 1;
    OpenAIChatItem chas_item = 2;
    string finish_reason = 3;
}
  
message Usage {
    int32 prompt_tokens = 1;
    int32 completion_tokens = 2;
    int32 total_tokens = 3;
}
  
message ChatCompletionRequest {
    string id = 1;
    string object = 2;
    int64 created = 3;
    repeated OpenAICompletionChoice choices = 4;
    Usage usage = 5;
}

completion.proto

syntax = "proto3";

package generate;

option go_package = "github.com/defenseunicorns/leapfrogai/pkg/client/generate";

// CompletionRequest is the payload to request completion
message CompletionRequest {
    string prompt = 1; // huggingface `inputs`
    optional string suffix = 2;
    optional int32 max_new_tokens = 3; // openai `max_tokens`
    optional float temperature = 4;
    optional int32 top_k = 5;
    optional float top_p = 6;
    optional bool do_sample = 7;
    optional int32 n = 8;
    optional int32 logprobs = 9;
    optional bool echo = 10;
    repeated string stop = 11;  // You can only represent Union[str, list] as a string. 
    optional float repetition_penalty = 12;
    optional float presence_penalty = 13;
    optional float frequence_penalty = 14;
    optional int32 best_of = 15;
    map<string, int32> logit_bias = 16;  // Maps are represented as a pair of a key type and a value type.
    optional bool return_full_text = 17;
    optional string truncate = 18;
    optional float typical_p = 19;
    optional bool watermark = 20; 
    optional int32 seed = 21;
    optional string user = 22;
}

enum CompletionFinishReason {
    STOP = 0;
    LENGTH = 1;
}

message CompletionChoice {
    string text = 1;
    int32 index = 2;
    CompletionFinishReason finish_reason = 3;
}

message CompletionUsage {
    int32 prompt_tokens = 1;
    int32 completion_tokens = 2;
    int32 total_tokens = 3;
}

// CompletionRespones are what's returned by the gRPC service
message CompletionResponse {
    repeated CompletionChoice choices = 1;
    CompletionUsage usage = 2;
}

service CompletionService {
    rpc Complete (CompletionRequest) returns (CompletionResponse);
    rpc CompleteStream (CompletionRequest) returns (stream CompletionResponse);
}