diff --git a/router/src/lib.rs b/router/src/lib.rs
index 0281c66d..3cff54c8 100644
--- a/router/src/lib.rs
+++ b/router/src/lib.rs
@@ -7,20 +7,11 @@ mod validation;
 #[cfg(feature = "kserve")]
 mod kserve;
 
-use infer::{Infer, InferError, InferStreamResponse};
-use queue::{Entry, Queue};
 use serde::{Deserialize, Serialize};
 use tracing::warn;
 use utoipa::ToSchema;
 use validation::Validation;
 
-/// Type alias for generation responses
-pub(crate) type GenerateStreamResponse = (
-    OwnedSemaphorePermit,
-    u32, // input_length
-    UnboundedReceiverStream<Result<InferStreamResponse, InferError>>,
-);
-
 #[derive(Clone, Deserialize, ToSchema)]
 pub(crate) struct VertexInstance {
     #[schema(example = "What is Deep Learning?")]
diff --git a/router/src/server.rs b/router/src/server.rs
index 46c199e8..0dec6929 100644
--- a/router/src/server.rs
+++ b/router/src/server.rs
@@ -4,7 +4,6 @@ use crate::infer::v2::SchedulerV2;
 use crate::infer::v3::SchedulerV3;
 use crate::infer::{HealthCheck, Scheduler};
 use crate::infer::{Infer, InferError, InferResponse, InferStreamResponse, ToolGrammar};
-use crate::health::Health;
 #[cfg(feature = "kserve")]
 use crate::kserve::{
     kerve_server_metadata, kserve_health_live, kserve_health_ready, kserve_model_infer,