Fix OSL issue due to PA not removing first 4 bytes from output

triton-inference-server · Jan 9, 2025 · 2441ba8 · 2441ba8
1 parent 14b4610
commit 2441ba8
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 4 deletions.
diff --git a/src/client_backend/triton_c_api/triton_loader.cc b/src/client_backend/triton_c_api/triton_loader.cc
@@ -1048,10 +1048,6 @@ TritonLoader::GetOutputs(
 
     std::string data_type{datatype};
     std::vector<uint8_t> data_copy;
-    if (data_type == "BYTES" && byte_size >= 4) {
-      base = static_cast<const uint8_t*>(base) + 4;
-      byte_size -= 4;
-    }
 
     if (memory_type == TRITONSERVER_MEMORY_GPU) {
       CUDARuntimeLibraryManager cuda_manager;

diff --git a/src/infer_context.cc b/src/infer_context.cc
@@ -220,6 +220,10 @@ InferContext::GetOutputs(const cb::InferResult& infer_result)
     std::vector<uint8_t> buf{};
     infer_result.RawData(requested_output->Name(), buf);
 
+    if (data_type == "BYTES" && buf.size() >= 4) {
+      buf.erase(buf.begin(), buf.begin() + 4);
+    }
+
     output.emplace(
         requested_output->Name(), RecordData(std::move(buf), data_type));
   }