Update tmux config

README.md

@@ -0,0 +1 @@
+# Rust Moxin: An AI LLM client built atop [Robius](https://g

moxin-protocol/src/open_ai.rs

@@ -125,3 +125,29 @@ pub struct ChunkChoiceData {
     pub finish_reason: Option<StopReason>,
     pub index: u32,
     pub delta: MessageData,
+    pub logprobs: Option<LogProbsData>,
+}
+
+#[derive(Clone, Debug, Serialize, Deserialize)]
+pub struct ChatResponseChunkData {
+    pub id: String,
+    pub choices: Vec<ChunkChoiceData>,
+    pub created: u32,
+    pub model: ModelID,
+    pub system_fingerprint: String,
+
+    #[serde(default = "response_chunk_object")]
+    pub object: String,
+}
+
+fn response_chunk_object() -> String {
+    "chat.completion.chunk".to_string()
+}
+
+#[derive(Clone, Debug)]
+pub enum ChatResponse {
+    // https://platform.openai.com/docs/api-reference/chat/object
+    ChatFinalResponseData(ChatResponseData),
+    // https://platform.openai.com/docs/api-reference/chat/streaming
+    ChatResponseChunk(ChatResponseChunkData),
+}

moxin-protocol/src/protocol.rs

@@ -0,0 +1,105 @@
+
+use crate::data::*;
+use crate::open_ai::*;
+use anyhow::Result;
+use std::sync::mpsc::Sender;
+
+#[derive(Clone, Debug)]
+pub enum FileDownloadResponse {
+    Progress(FileID, f32),
+    Completed(DownloadedFile),
+}
+
+#[derive(Clone, Debug)]
+pub enum ContextOverflowPolicy {
+    StopAtLimit,
+    TruncateMiddle,
+    TruncatePastMessages,
+}
+
+#[derive(Clone, Debug)]
+pub enum GPULayers {
+    Specific(u32),
+    Max,
+}
+
+#[derive(Clone, Debug)]
+pub struct LoadModelOptions {
+    pub prompt_template: Option<String>,
+    pub gpu_layers: GPULayers,
+    pub use_mlock: bool,
+    pub n_batch: u32,
+    pub n_ctx: u32,
+    pub rope_freq_scale: f32,
+    pub rope_freq_base: f32,
+
+    // TBD Not really sure if this is something backend manages or if it is matter of
+    // the client (if it is done by tweaking the JSON payload for the chat completition)
+    pub context_overflow_policy: ContextOverflowPolicy,
+}
+
+#[derive(Clone, Debug)]
+pub struct LoadedModelInfo {
+    pub file_id: FileID,
+    pub model_id: ModelID,
+
+    // JSON formatted string with the model information. See "Model Inspector" in LMStudio.
+    pub information: String,
+}
+
+#[derive(Clone, Debug)]
+pub struct ModelResourcesInfo {
+    pub ram_usage: f32,
+    pub cpu_usage: f32,
+}
+
+#[derive(Clone, Debug)]
+pub enum LoadModelResponse {
+    Progress(FileID, f32),
+    Completed(LoadedModelInfo),
+    ModelResourcesUsage(ModelResourcesInfo),
+}
+
+#[derive(Clone, Debug)]
+pub struct LocalServerConfig {
+    pub port: u16,
+    pub cors: bool,
+    pub request_queuing: bool,
+    pub verbose_server_logs: bool,
+    pub apply_prompt_formatting: bool,
+}
+
+#[derive(Clone, Debug)]
+pub enum LocalServerResponse {
+    Started,
+    Log(String),
+}
+
+#[derive(Clone, Debug)]
+pub enum Command {
+    GetFeaturedModels(Sender<Result<Vec<Model>>>),
+
+    // The argument is a string with the keywords to search for.
+    SearchModels(String, Sender<Result<Vec<Model>>>),
+
+    DownloadFile(FileID, Sender<Result<FileDownloadResponse>>),
+    PauseDownload(FileID, Sender<Result<()>>),
+    CancelDownload(FileID, Sender<Result<()>>),
+    DeleteFile(FileID, Sender<Result<()>>),
+
+    GetCurrentDownloads(Sender<Result<Vec<PendingDownload>>>),
+    GetDownloadedFiles(Sender<Result<Vec<DownloadedFile>>>),
+
+    LoadModel(FileID, LoadModelOptions, Sender<Result<LoadModelResponse>>),
+
+    // Eject currently loaded model, if any is provided
+    EjectModel(Sender<Result<()>>),
+
+    Chat(ChatRequestData, Sender<Result<ChatResponse>>),
+    StopChatCompletion(Sender<Result<()>>),
+
+    // Command to start a local server to interact with chat models
+    StartLocalServer(LocalServerConfig, Sender<Result<LocalServerResponse>>),
+    // Command to stop the local server
+    StopLocalServer(Sender<Result<()>>),
+}