[model-gateway] Fix logging module name, parse endpoint context, and tokenizer factory (#15782)

sgl-model-gateway/src/middleware.rs

@@ -289,7 +289,7 @@ impl<B> MakeSpan<B> for RequestSpan {
             status_code = Empty,
             latency = Empty,
             error = Empty,
-            module = "sglang::router_rs"
+            module = "sgl_model_gateway"
         )
     }
 }

sgl-model-gateway/src/routers/http/router.rs

@@ -35,14 +35,13 @@ use crate::{
         completion::CompletionRequest,
         embedding::EmbeddingRequest,
         generate::GenerateRequest,
-        parser::{ParseFunctionCallRequest, SeparateReasoningRequest},
         rerank::{RerankRequest, RerankResponse, RerankResult},
         responses::{ResponsesGetParams, ResponsesRequest},
     },
     routers::{
         error::{self, extract_error_code_from_response},
         grpc::utils::{error_type_from_status, route_to_endpoint},
-        header_utils, parse, RouterTrait,
+        header_utils, RouterTrait,
     },
 };
 
@@ -54,7 +53,6 @@ pub struct Router {
     dp_aware: bool,
     enable_igw: bool,
     retry_config: RetryConfig,
-    context: Option<Arc<AppContext>>,
 }
 
 impl std::fmt::Debug for Router {
@@ -66,7 +64,6 @@ impl std::fmt::Debug for Router {
             .field("dp_aware", &self.dp_aware)
             .field("enable_igw", &self.enable_igw)
             .field("retry_config", &self.retry_config)
-            .field("context", &"<AppContext>")
             .finish()
     }
 }
@@ -81,7 +78,6 @@ impl Router {
             dp_aware: ctx.router_config.dp_aware,
             enable_igw: ctx.router_config.enable_igw,
             retry_config: ctx.router_config.effective_retry_config(),
-            context: Some(ctx.clone()),
         })
     }
 
@@ -817,14 +813,6 @@ impl RouterTrait for Router {
         }
     }
 
-    async fn parse_function_call(&self, req: &ParseFunctionCallRequest) -> Response {
-        parse::parse_function_call(self.context.as_ref(), req).await
-    }
-
-    async fn parse_reasoning(&self, req: &SeparateReasoningRequest) -> Response {
-        parse::parse_reasoning(self.context.as_ref(), req).await
-    }
-
     fn router_type(&self) -> &'static str {
         "regular"
     }
@@ -859,7 +847,6 @@ mod tests {
             client: Client::new(),
             retry_config: RetryConfig::default(),
             enable_igw: false,
-            context: None,
         }
     }
 

sgl-model-gateway/src/routers/mod.rs

@@ -16,7 +16,6 @@ use crate::protocols::{
     completion::CompletionRequest,
     embedding::EmbeddingRequest,
     generate::GenerateRequest,
-    parser::{ParseFunctionCallRequest, SeparateReasoningRequest},
     rerank::RerankRequest,
     responses::{ResponsesGetParams, ResponsesRequest},
 };
@@ -194,16 +193,6 @@ pub trait RouterTrait: Send + Sync + Debug {
         (StatusCode::NOT_IMPLEMENTED, "Rerank not implemented").into_response()
     }
 
-    /// Parse function calls from text
-    async fn parse_function_call(&self, req: &ParseFunctionCallRequest) -> Response {
-        parse::parse_function_call(None, req).await
-    }
-
-    /// Separate reasoning from normal text
-    async fn parse_reasoning(&self, req: &SeparateReasoningRequest) -> Response {
-        parse::parse_reasoning(None, req).await
-    }
-
     /// Get router type name
     fn router_type(&self) -> &'static str;
 

sgl-model-gateway/src/routers/parse/handlers.rs

@@ -28,13 +28,9 @@ fn error_response(status: StatusCode, message: &str) -> Response {
 
 /// Parse function calls from model output text
 pub async fn parse_function_call(
-    context: Option<&Arc<AppContext>>,
+    ctx: &Arc<AppContext>,
     req: &ParseFunctionCallRequest,
 ) -> Response {
-    let Some(ctx) = context else {
-        return error_response(StatusCode::SERVICE_UNAVAILABLE, "Context not initialized");
-    };
-
     let Some(factory) = &ctx.tool_parser_factory else {
         return error_response(
             StatusCode::SERVICE_UNAVAILABLE,
@@ -71,14 +67,7 @@ pub async fn parse_function_call(
 }
 
 /// Parse and separate reasoning from normal text
-pub async fn parse_reasoning(
-    context: Option<&Arc<AppContext>>,
-    req: &SeparateReasoningRequest,
-) -> Response {
-    let Some(ctx) = context else {
-        return error_response(StatusCode::SERVICE_UNAVAILABLE, "Context not initialized");
-    };
-
+pub async fn parse_reasoning(ctx: &Arc<AppContext>, req: &SeparateReasoningRequest) -> Response {
     let Some(factory) = &ctx.reasoning_parser_factory else {
         return error_response(
             StatusCode::SERVICE_UNAVAILABLE,

sgl-model-gateway/src/server.rs

@@ -50,7 +50,7 @@ use crate::{
         validated::ValidatedJson,
         worker_spec::{WorkerConfigRequest, WorkerUpdateRequest},
     },
-    routers::{conversations, router_manager::RouterManager, tokenize, RouterTrait},
+    routers::{conversations, parse, router_manager::RouterManager, tokenize, RouterTrait},
     service_discovery::{start_service_discovery, ServiceDiscoveryConfig},
     wasm::route::{add_wasm_module, list_wasm_modules, remove_wasm_module},
     workflow::{LoggingSubscriber, WorkflowEngine},
@@ -67,14 +67,14 @@ async fn parse_function_call(
     State(state): State<Arc<AppState>>,
     Json(req): Json<ParseFunctionCallRequest>,
 ) -> Response {
-    state.router.parse_function_call(&req).await
+    parse::parse_function_call(&state.context, &req).await
 }
 
 async fn parse_reasoning(
     State(state): State<Arc<AppState>>,
     Json(req): Json<SeparateReasoningRequest>,
 ) -> Response {
-    state.router.parse_reasoning(&req).await
+    parse::parse_reasoning(&state.context, &req).await
 }
 
 async fn sink_handler() -> Response {

sgl-model-gateway/src/tokenizer/factory.rs

@@ -254,14 +254,27 @@ pub async fn create_tokenizer_async_with_chat_template(
     }
 
     // Check if it's a GPT model name that should use Tiktoken
-    if model_name_or_path.contains("gpt-")
+    // Only match specific OpenAI model patterns to avoid catching HuggingFace models like "openai/gpt-oss-20b"
+    if model_name_or_path.contains("gpt-4")
+        || model_name_or_path.contains("gpt-3.5")
+        || model_name_or_path.contains("gpt-3")
+        || model_name_or_path.contains("turbo")
         || model_name_or_path.contains("davinci")
         || model_name_or_path.contains("curie")
         || model_name_or_path.contains("babbage")
         || model_name_or_path.contains("ada")
+        || model_name_or_path.contains("codex")
     {
-        let tokenizer = TiktokenTokenizer::from_model_name(model_name_or_path)?;
-        return Ok(Arc::new(tokenizer));
+        // Try tiktoken first, but fall back to HuggingFace if it fails
+        match TiktokenTokenizer::from_model_name(model_name_or_path) {
+            Ok(tokenizer) => return Ok(Arc::new(tokenizer)),
+            Err(e) => {
+                debug!(
+                    "Tiktoken failed for '{}': {}, falling back to HuggingFace",
+                    model_name_or_path, e
+                );
+            }
+        }
     }
 
     // Try to download tokenizer files from HuggingFace