Add the llama client types and the chat request builder

Implemented-By: Laguna S 2.1 (OpenCode)

crates/loopd/src/lib.rs

@@ -2,4 +2,5 @@
 
 pub mod config;
 pub mod http;
+pub mod llama;
 pub mod sse;

crates/loopd/src/llama/mod.rs

@@ -0,0 +1,145 @@
+//! The llama inference server's chat-completions types and the request builder.
+
+pub mod request;
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum ChatMessage {
+    System {
+        content: String,
+    },
+    User {
+        content: String,
+    },
+    Assistant {
+        content: Option<String>,
+        reasoning_content: Option<String>,
+        tool_calls: Vec<proto::ToolCall>,
+    },
+    Tool {
+        tool_call_id: String,
+        content: String,
+    },
+}
+
+#[derive(Debug, Clone, PartialEq)]
+pub struct ToolSchema {
+    pub name: String,
+    pub description: String,
+    pub parameters: serde_json::Value,
+}
+
+#[derive(Debug, Clone, PartialEq)]
+pub struct ChatRequest {
+    pub slot: u32,
+    pub messages: Vec<ChatMessage>,
+    pub tools: Vec<ToolSchema>,
+    pub thinking: bool,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum ChatEvent {
+    Queued {
+        ahead: usize,
+    },
+    Waiting {
+        slot_busy: bool,
+    },
+    Progress {
+        total: u64,
+        cache: u64,
+        processed: u64,
+    },
+    Reasoning(String),
+    Content(String),
+    ToolCallDelta {
+        index: u32,
+        id: Option<String>,
+        name: Option<String>,
+        arguments: String,
+    },
+    ThinkingCapped {
+        tokens: u64,
+    },
+    Retrying {
+        attempt: u32,
+        after_ms: u64,
+        error: String,
+    },
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum FinishReason {
+    Stop,
+    ToolCalls,
+    Length,
+}
+
+/// The three counts loopd uses. The server sends more fields beside them; they are ignored.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Default, serde::Deserialize)]
+pub struct Timings {
+    pub cache_n: u64,
+    pub prompt_n: u64,
+    pub predicted_n: u64,
+}
+
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct Completion {
+    pub id: String,
+    pub content: Option<String>,
+    pub reasoning_content: Option<String>,
+    pub tool_calls: Vec<proto::ToolCall>,
+    pub finish_reason: FinishReason,
+    pub timings: Timings,
+    pub reasoning_tokens: u64,
+    pub thinking_capped: bool,
+}
+
+#[derive(Debug)]
+pub enum InferError {
+    Busy,
+    Connect(std::io::Error),
+    WaitTimeout,
+    LoadTimeout,
+    Stalled,
+    StreamClosedEarly,
+    ThinkingOverrun,
+    Http { status: u16, body: String },
+    Protocol(String),
+}
+
+impl std::fmt::Display for InferError {
+    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        match self {
+            InferError::Busy => write!(f, "the slot is busy"),
+            InferError::Connect(err) => {
+                write!(f, "could not connect to the inference server: {err}")
+            }
+            InferError::WaitTimeout => write!(f, "timed out waiting for the slot to free up"),
+            InferError::LoadTimeout => write!(f, "timed out waiting for the model to load"),
+            InferError::Stalled => write!(f, "the stream stalled"),
+            InferError::StreamClosedEarly => {
+                write!(f, "the stream closed before the response was complete")
+            }
+            InferError::ThinkingOverrun => write!(f, "thinking ran over its token budget"),
+            InferError::Http { status, body } => {
+                write!(f, "the server responded with {status}: {body}")
+            }
+            InferError::Protocol(what) => write!(f, "the server spoke the protocol wrong: {what}"),
+        }
+    }
+}
+
+impl std::error::Error for InferError {}
+
+pub struct Client {
+    pub(crate) cfg: crate::config::Config,
+}
+
+impl Client {
+    pub fn new(cfg: crate::config::Config) -> Self {
+        Self { cfg }
+    }
+    pub fn config(&self) -> &crate::config::Config {
+        &self.cfg
+    }
+}

crates/loopd/src/llama/request.rs

@@ -0,0 +1,177 @@
+//! Building the chat request body the inference server expects.
+
+use serde::Serialize;
+
+use super::{ChatMessage, ToolSchema};
+
+/// The request body as JSON text. The same inputs always give the same bytes.
+pub fn build_body(
+    cfg: &crate::config::Config,
+    req: &super::ChatRequest,
+) -> Result<String, serde_json::Error> {
+    let messages = serialize_messages(&req.messages)?;
+    let tools = req.tools.iter().map(serialized_tool).collect::<Vec<_>>();
+    let body = RequestBody {
+        model: cfg.infer.model.clone(),
+        messages,
+        tools,
+        id_slot: req.slot,
+        cache_prompt: true,
+        stream: true,
+        return_progress: true,
+        timings_per_token: true,
+        reasoning_control: true,
+        max_tokens: cfg.limits.max_tokens,
+        temperature: cfg.sampling.temperature,
+        top_p: cfg.sampling.top_p,
+        top_k: cfg.sampling.top_k,
+        chat_template_kwargs: ChatTemplateKwargs {
+            enable_thinking: req.thinking,
+        },
+    };
+    serde_json::to_string(&body)
+}
+
+fn serialize_messages(
+    messages: &[ChatMessage],
+) -> Result<Vec<serde_json::Value>, serde_json::Error> {
+    let mut out = Vec::with_capacity(messages.len());
+    for message in messages {
+        out.push(serialize_message(message)?);
+    }
+    Ok(out)
+}
+
+fn serialize_message(message: &ChatMessage) -> Result<serde_json::Value, serde_json::Error> {
+    match message {
+        ChatMessage::System { content } => serde_json::to_value(SerializedSystem {
+            role: "system",
+            content: content.clone(),
+        }),
+        ChatMessage::User { content } => serde_json::to_value(SerializedUser {
+            role: "user",
+            content: content.clone(),
+        }),
+        ChatMessage::Tool {
+            tool_call_id,
+            content,
+        } => serde_json::to_value(SerializedToolMessage {
+            role: "tool",
+            tool_call_id: tool_call_id.clone(),
+            content: content.clone(),
+        }),
+        ChatMessage::Assistant {
+            content,
+            reasoning_content,
+            tool_calls,
+        } => serde_json::to_value(SerializedAssistant {
+            role: "assistant",
+            content: content.clone().unwrap_or_default(),
+            reasoning_content: reasoning_content.clone(),
+            tool_calls: tool_calls.iter().map(serialized_call).collect(),
+        }),
+    }
+}
+
+fn serialized_call(call: &proto::ToolCall) -> SerializedToolCall {
+    SerializedToolCall {
+        id: call.id.clone(),
+        kind: "function",
+        function: CallFunction {
+            name: call.name.clone(),
+            arguments: call.arguments.clone(),
+        },
+    }
+}
+
+fn serialized_tool(tool: &ToolSchema) -> SerializedTool {
+    SerializedTool {
+        kind: "function",
+        function: ToolFunction {
+            name: tool.name.clone(),
+            description: tool.description.clone(),
+            parameters: tool.parameters.clone(),
+        },
+    }
+}
+
+#[derive(Serialize)]
+struct RequestBody {
+    model: String,
+    messages: Vec<serde_json::Value>,
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    tools: Vec<SerializedTool>,
+    id_slot: u32,
+    cache_prompt: bool,
+    stream: bool,
+    return_progress: bool,
+    timings_per_token: bool,
+    reasoning_control: bool,
+    max_tokens: u64,
+    temperature: f64,
+    top_p: f64,
+    top_k: u32,
+    chat_template_kwargs: ChatTemplateKwargs,
+}
+
+#[derive(Serialize)]
+struct ChatTemplateKwargs {
+    enable_thinking: bool,
+}
+
+#[derive(Serialize)]
+struct SerializedSystem {
+    role: &'static str,
+    content: String,
+}
+
+#[derive(Serialize)]
+struct SerializedUser {
+    role: &'static str,
+    content: String,
+}
+
+#[derive(Serialize)]
+struct SerializedToolMessage {
+    role: &'static str,
+    tool_call_id: String,
+    content: String,
+}
+
+#[derive(Serialize)]
+struct SerializedAssistant {
+    role: &'static str,
+    content: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    reasoning_content: Option<String>,
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    tool_calls: Vec<SerializedToolCall>,
+}
+
+#[derive(Serialize)]
+struct SerializedToolCall {
+    id: String,
+    #[serde(rename = "type")]
+    kind: &'static str,
+    function: CallFunction,
+}
+
+#[derive(Serialize)]
+struct CallFunction {
+    name: String,
+    arguments: String,
+}
+
+#[derive(Serialize)]
+struct SerializedTool {
+    #[serde(rename = "type")]
+    kind: &'static str,
+    function: ToolFunction,
+}
+
+#[derive(Serialize)]
+struct ToolFunction {
+    name: String,
+    description: String,
+    parameters: serde_json::Value,
+}