feat: add unload llm route

src/api/routes/v1/llm.rs

@@ -106,46 +106,52 @@ pub async fn load_model(
     }))
 }
 
-// #[utoipa::path(
+#[utoipa::path(
-//     delete,
+    post,
-//     path = "/models/{model}/load",
+    path = "/models/{model}/unload",
-//     tag = "models",
+    tag = "models",
-//     params(
+    params(
-//         ("model" = String, Path, description = "Name of the model to unload from memory (e.g. 'llama3')")
+        ("model" = String, Path, description = "Name of the model to unload from memory (e.g. 'llama3')")
-//     ),
+    ),
-//     responses(
+    responses(
-//         (
+        (
-//             status = 200,
+            status = 200,
-//             description = "Model successfully unloaded from memory",
+            description = "Model successfully unloaded from memory",
-//             body = api::types::UnloadModelResponse,
+            body = api::types::ApiUnloadModelResponse,
-//             content_type = "application/json",
+            content_type = "application/json",
-//         ),
+        ),
-//         (
+        (
-//             status = 404,
+            status = 404,
-//             description = "Model not found locally",
+            description = "Model not found locally",
-//             body = api::errors::ErrorResponse,
+            body = api::errors::ErrorResponse,
-//             example = json!({ "error": "model 'llama3' not found — run `ollama pull llama3`" })
+            example = json!({ "error": "model 'llama3' not found — run `ollama pull llama3`" })
-//         ),
+        ),
-//         (
+        (
-//             status = 500,
+            status = 500,
-//             description = "Internal server error (Ollama or network failure)",
+            description = "Internal server error (Ollama or network failure)",
-//             body = api::errors::ErrorResponse,
+            body = api::errors::ErrorResponse,
-//             example = json!({ "error": "connection refused" })
+            example = json!({ "error": "connection refused" })
-//         )
+        )
-//     )
+    )
-// )]
+)]
-// pub async fn unload_model(
+pub async fn unload_model(
-//     State(state): State<AppState>,
+    State(state): State<SharedState>,
-//     Path(model): Path<String>,
+    Path(model): Path<String>,
-// ) -> Result<Json<api::types::UnloadModelResponse>, (axum::http::StatusCode, String)> {
+    Json(body): Json<api::types::ApiUnloadModelRequest>,
-//     let response = state
+) -> Result<Json<api::types::ApiUnloadModelResponse>, api::errors::ApiError> {
-//         .ollama
+    let response = state
-//         .unload_model(&model)
+        .chat_service
-//         .await
+        .unload_model(crate::core::llm::models::UnloadModelRequest {
-//         .map_err(into_http_response)?;
+            model,
+            keep_alive: body.keep_alive.clone(),
+        })
+        .await?;
 
-//     Ok(Json(response))
+    Ok(Json(api::types::ApiUnloadModelResponse {
-// }
+        model: response.model,
+        status: "unloaded".to_string(),
+    }))
+}
 
 #[utoipa::path(
     post,

src/api/routes/v1/mod.rs

@@ -22,6 +22,7 @@ fn llm_router() -> Router<SharedState> {
         .route("/completions", post(llm::completions))
         .route("/chat/completions", post(llm::chat_completions))
         .route("/models/{model}/load", post(llm::load_model))
+        .route("/models/{model}/unload", post(llm::unload_model))
 }
 
 fn keys_router() -> Router<SharedState> {

src/api/types.rs

@@ -47,6 +47,11 @@ pub struct ApiUnloadModelResponse {
     pub status: String,
 }
 
+#[derive(Debug, Serialize, Deserialize, ToSchema)]
+pub struct ApiUnloadModelRequest {
+    pub keep_alive: String,
+}
+
 // ------ Completions ------
 
 #[derive(Debug, Deserialize, Serialize, ToSchema, Default)]

src/core/llm/models.rs

@@ -34,3 +34,14 @@ pub struct LoadModelRequest {
 pub struct LoadModelResponse {
     pub model: String,
 }
+
+#[derive(Debug, Clone)]
+pub struct UnloadModelRequest {
+    pub model: String,
+    pub keep_alive: String,
+}
+
+#[derive(Debug, Clone)]
+pub struct UnloadModelResponse {
+    pub model: String,
+}

src/services/chat_service.rs

@@ -50,6 +50,23 @@ impl ChatService {
         Ok(crate::core::llm::models::LoadModelResponse { model: body.model })
     }
 
+    pub async fn unload_model(
+        &self,
+        body: crate::core::llm::models::UnloadModelRequest,
+    ) -> Result<crate::core::llm::models::UnloadModelResponse, ServiceError> {
+        let b = crate::providers::ollama::types::OllamaGenerateRequest {
+            model: body.model.clone(),
+            prompt: "unload".to_string(),
+            stream: false,
+            keep_alive: body.keep_alive,
+            options: None,
+        };
+
+        self.ollama.completions(&b).await?;
+
+        Ok(crate::core::llm::models::UnloadModelResponse { model: body.model })
+    }
+
     pub async fn complete(
         &self,
         body: core::llm::completions::CompletionRequest,