diff --git a/backends/trtllm/csrc/backend.cpp b/backends/trtllm/csrc/backend.cpp
index 5e52261e..b50044d8 100644
--- a/backends/trtllm/csrc/backend.cpp
+++ b/backends/trtllm/csrc/backend.cpp
@@ -7,7 +7,7 @@
 #include "hardware.hpp"
 
 namespace huggingface::tgi::backends::trtllm {
-    constexpr tle::ParallelConfig backend_workspace_t::parallel_config() const {
+    tle::ParallelConfig backend_workspace_t::parallel_config() const {
         // Single engine (TP = PP = 1) -> using leader mode (no MPI involved)
         const auto world_size = config_["/pretrained_config/mapping/world_size"_json_pointer].get<size_t>();
 
diff --git a/backends/trtllm/csrc/backend.hpp b/backends/trtllm/csrc/backend.hpp
index 5a08b2d4..61aa11d0 100644
--- a/backends/trtllm/csrc/backend.hpp
+++ b/backends/trtllm/csrc/backend.hpp
@@ -127,7 +127,7 @@ namespace huggingface::tgi::backends::trtllm {
          * to initialize `tensorrt_llm::executor::Executor` with multi-instance communication information
          * @return `tensorrt_llm::executor::ParallelConfig` instance
          */
-        [[nodiscard]] constexpr tle::ParallelConfig parallel_config() const;
+        [[nodiscard]] tle::ParallelConfig parallel_config() const;
 
         /**
          * Factory method returning new `tensorrt_llm::executor::ExecutorConfig` instance used