[router] improve reasoning parser lock and reduce req cloning (#11336)
This commit is contained in:
@@ -129,7 +129,7 @@ impl GrpcRouter {
|
||||
// Use pipeline for ALL requests (streaming and non-streaming)
|
||||
self.pipeline
|
||||
.execute_chat(
|
||||
body.clone(),
|
||||
Arc::new(body.clone()),
|
||||
headers.cloned(),
|
||||
model_id.map(|s| s.to_string()),
|
||||
self.shared_components.clone(),
|
||||
@@ -149,7 +149,7 @@ impl GrpcRouter {
|
||||
// Use pipeline for ALL requests (streaming and non-streaming)
|
||||
self.pipeline
|
||||
.execute_generate(
|
||||
body.clone(),
|
||||
Arc::new(body.clone()),
|
||||
headers.cloned(),
|
||||
model_id.map(|s| s.to_string()),
|
||||
self.shared_components.clone(),
|
||||
|
||||
Reference in New Issue
Block a user