From 4e613a55cf38f11a54160cf8eead4d8947a5115b Mon Sep 17 00:00:00 2001
From: Iman Tabrizian <itabrizian@nvidia.com>
Date: Mon, 11 Sep 2023 15:48:20 -0400
Subject: [PATCH] Fix null requests

---
 src/backend_model_instance.cc | 3 ++-
 src/infer_request.cc          | 1 +
 2 files changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/backend_model_instance.cc b/src/backend_model_instance.cc
index cf42b5bd7..526c558f9 100644
--- a/src/backend_model_instance.cc
+++ b/src/backend_model_instance.cc
@@ -633,7 +633,8 @@ TritonModelInstance::WarmUp()
             reinterpret_cast<TRITONBACKEND_Request*>(request.get()));
 
         // For warmup requests we need to manually set ResponseFactory
-        // since they don't run `PrepareForInference`.
+        // since they modify the callback after PrepareForInference has
+        // been called.
         request->SetResponseFactory();
       }
 
diff --git a/src/infer_request.cc b/src/infer_request.cc
index 438e3d272..57828e8dc 100644
--- a/src/infer_request.cc
+++ b/src/infer_request.cc
@@ -590,6 +590,7 @@ InferenceRequest::CopyAsNull(const InferenceRequest& from)
   lrequest->SetResponseCallback(
       &null_allocator, nullptr, NullResponseComplete, nullptr);
   lrequest->SetReleaseCallback(NullRequestComplete, nullptr);
+  lrequest->SetResponseFactory();
 
   // Must normalize inputs here...
   for (auto& pr : lrequest->original_inputs_) {