fix: more

Cargo.lock

@@ -709,6 +709,9 @@ name = "ecow"
 version = "0.3.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2dca6141d10838e13f9b166c190254a3f9b5d8379521f9b7510810061cef17a4"
+dependencies = [
+ "serde",
+]
 
 [[package]]
 name = "either"

Cargo.toml

@@ -8,7 +8,7 @@ async-channel = "2.5.0"
 axum = { version = "0.8.9", features = ["http2", "macros", "multipart", "ws"] }
 axum-extra = { version = "0.12.6", features = ["typed-header", "typed-routing"] }
 clap = { version = "4.6.7", features = ["derive", "env"] }
-ecow = "0.3.1"
+ecow = { version = "0.3.1", features = ["serde"] }
 enum-map = { version = "3.1.0", features = ["serde"] }
 enumset = { version = "1.1.14", features = ["serde"] }
 futures = "0.3.34"

src/main.rs

@@ -12,10 +12,10 @@ mod server;
 
 #[derive(Parser, Debug)]
 struct AppArgs {
-    #[arg(long, env)]
+    #[arg(long, env, default_value_t = [127, 0, 0, 1].into())]
     ip: IpAddr,
 
-    #[arg(long, env)]
+    #[arg(long, env, default_value_t = 3126)]
     port: u16,
 
     #[arg(long)]
@@ -106,6 +106,7 @@ async fn main() {
     };
 
     join!(stream_responses, async move {
+        tracing::info!(?addr, "listening");
         axum::serve(listener, app).await.expect("TODO");
     });
 }

src/server.rs

@@ -45,8 +45,8 @@ pub struct AppState {
 }
 
 #[derive(Deserialize, TypedPath)]
-#[typed_path("/chat/completions")]
+#[typed_path("/v1/chat/completions")]
-struct ChatCompletionsPath;
+struct V1ChatCompletionsPath;
 
 #[derive(Debug, Clone)]
 enum Provider {
@@ -59,6 +59,7 @@ enum Provider {
 }
 
 #[derive(Debug, Serialize, Deserialize, Clone)]
+#[serde(rename_all = "lowercase")]
 enum Shortcut {
     XS,
     S,
@@ -142,8 +143,8 @@ impl Shortcut {
 }
 
 fn take_random_by_swapping<T>(mut vec1: Vec1<T>, mut rng: impl Rng) -> (T, Option<Vec1<T>>) {
-    let end = vec1.len().get();
+    let end = vec1.len().get() - 1;
-    let index = rng.random_range(0..end);
+    let index = rng.random_range(0..=end);
 
     vec1.swap(index, end);
     let (body, element) = vec1.into_rtail_and_head();
@@ -206,7 +207,7 @@ impl IntoResponse for ChatCompletionError {
 #[axum::debug_handler]
 #[tracing::instrument(skip(state, request))]
 async fn create_chat_completion(
-    _: ChatCompletionsPath,
+    _: V1ChatCompletionsPath,
     State(state): State<AppState>,
     Json(request): Json<openai::chat::completion::CreateChatCompletionRequest<Shortcut>>,
 ) -> Result<ChatCompletionResponse, ChatCompletionError> {
@@ -293,6 +294,8 @@ async fn create_chat_completion(
             model_upstreams.push_back((model.clone(), upstreams));
         }
 
+        tracing::info!(?model, ?upstream);
+
         let openai::chat::completion::CreateChatCompletionRequest {
             model: _shortcut,
             messages,
@@ -375,18 +378,25 @@ async fn create_chat_completion(
 
         let client = reqwest::Client::new();
 
-        let mut upstream_request = client.post({
+        let mut upstream_request = client
-            let mut url = upstream.base_url.clone();
+            .post({
+                let mut url = upstream.base_url;
 
-            {
+                {
-                let mut segments = url.path_segments_mut().expect("TODO");
+                    let mut path_segments = url.path_segments_mut().expect("TODO");
 
-                segments.push("chat");
+                    let s = V1ChatCompletionsPath.to_string();
-                segments.push("completions");
+                    let s = s.strip_prefix("/").unwrap_or(&s);
-            }
+                    let s = s.strip_suffix("/").unwrap_or(&s);
 
-            url
+                    path_segments.extend(s.split("/"));
-        });
+                }
+
+                tracing::info!(?url);
+
+                url
+            })
+            .json(&request);
 
         if let Some(api_key) = upstream.api_key {
             let authorization =
@@ -410,12 +420,13 @@ async fn create_chat_completion(
             }
         };
 
-        let upstream_response = match upstream_response.error_for_status() {
+        if let Err(upstream_error) = upstream_response.error_for_status_ref() {
-            Ok(upstream_response) => upstream_response,
+            tracing::error!(?upstream_error);
-            Err(upstream_error) => {
+            match upstream_response.text().await {
-                tracing::error!(?upstream_error);
+                Ok(body) => tracing::error!(?body),
-                continue;
+                Err(body_error) => tracing::error!(?body_error),
-            }
+            };
+            continue;
         };
 
         if request.stream.unwrap_or(false) {