diff --git a/examples/arm/executor_runner/arm_executor_runner.cpp b/examples/arm/executor_runner/arm_executor_runner.cpp
index ea54c9ca605..d56710e27ad 100644
--- a/examples/arm/executor_runner/arm_executor_runner.cpp
+++ b/examples/arm/executor_runner/arm_executor_runner.cpp
@@ -763,13 +763,7 @@ void log_mem_status(const RunnerContext& ctx) {
     ET_LOG(Info, "method_allocator_executor: %zu bytes", executor_memsize);
   }
   if (ctx.temp_allocator->size() > 0) {
-    ET_LOG(
-        Info,
-        "peak_temp_allocator:       %zu / %zu free: %zu ( used: %zu %% ) ",
-        ctx.temp_allocator->peak_used(),
-        ctx.temp_allocator->size(),
-        ctx.temp_allocator->free_size(),
-        100 * ctx.temp_allocator->peak_used() / ctx.temp_allocator->size());
+    ET_LOG(Info, "temp_allocator:            %zu", ctx.temp_allocator->size());
   }
 }
 
@@ -927,15 +921,20 @@ void verify_result(RunnerContext& ctx, const void* model_pte) {
 void run_model(RunnerContext& ctx, const void* model_pte) {
   Error status;
   ET_LOG(Info, "Starting running %d inferences...", num_inferences);
-
   int n = 0;
   StartMeasurements();
   for (n = 0; n < num_inferences; n++) {
+    ET_LOG(Debug, "Running inference number %d", n);
     // Run the model.
     status = ctx.method.value()->execute();
     if (status != Error::Ok) {
       break;
     }
+    // Reset the temporary allocator holding the scratch buffer between
+    // inferences. We want to reuse the temp_allocator between inferences of the
+    // same Ethos-U custom delegate, not allocate memory with every new
+    // inference.
+    ctx.temp_allocator.reset(temp_allocation_pool_size, temp_allocation_pool);
   }
   StopMeasurements(n);
 
diff --git a/examples/arm/executor_runner/arm_memory_allocator.cpp b/examples/arm/executor_runner/arm_memory_allocator.cpp
index e22439a239d..6b627625ae1 100644
--- a/examples/arm/executor_runner/arm_memory_allocator.cpp
+++ b/examples/arm/executor_runner/arm_memory_allocator.cpp
@@ -7,7 +7,7 @@
 #include "arm_memory_allocator.h"
 
 ArmMemoryAllocator::ArmMemoryAllocator(uint32_t size, uint8_t* base_address)
-    : MemoryAllocator(size, base_address), used_(0), peak_used_(0) {}
+    : MemoryAllocator(size, base_address), used_(0) {}
 
 void* ArmMemoryAllocator::allocate(size_t size, size_t alignment) {
   void* ret = executorch::runtime::MemoryAllocator::allocate(size, alignment);
@@ -22,8 +22,6 @@ void* ArmMemoryAllocator::allocate(size_t size, size_t alignment) {
     } else {
       used_ = (used_ | (alignment - 1)) + 1 + size;
     }
-    if (used_ > peak_used_)
-      peak_used_ = used_;
   }
   return ret;
 }
@@ -32,10 +30,6 @@ size_t ArmMemoryAllocator::used_size() const {
   return used_;
 }
 
-size_t ArmMemoryAllocator::peak_used() const {
-  return peak_used_;
-}
-
 size_t ArmMemoryAllocator::free_size() const {
   return executorch::runtime::MemoryAllocator::size() - used_;
 }
diff --git a/examples/arm/executor_runner/arm_memory_allocator.h b/examples/arm/executor_runner/arm_memory_allocator.h
index f7e8939c655..1d7bbdecb4c 100644
--- a/examples/arm/executor_runner/arm_memory_allocator.h
+++ b/examples/arm/executor_runner/arm_memory_allocator.h
@@ -21,15 +21,10 @@ class ArmMemoryAllocator : public executorch::runtime::MemoryAllocator {
   // Returns the used size of the allocator's memory buffer.
   size_t used_size() const;
 
-  // Returns the peak memory usage of the allocator's memory buffer
-  // Peak usage is useful when doing multiple allocations & resets
-  size_t peak_used() const;
-
   // Returns the free size of the allocator's memory buffer.
   size_t free_size() const;
   void reset();
 
  private:
   size_t used_;
-  size_t peak_used_;
 };