Disable CPU arena allocator for ONNX

The arena memory allocator pre-allocates excessive of memory up front.
Disabling matches the existing configuration in ONNX integration for backend.

1 files changed, 1 insertions, 0 deletions

diff --git a/model-integration/src/main/java/ai/vespa/modelintegration/evaluator/OnnxEvaluatorOptions.java b/model-integration/src/main/java/ai/vespa/modelintegration/evaluator/OnnxEvaluatorOptions.java
index 6dd2c5b05af..cefafc3654b 100644
--- a/model-integration/src/main/java/ai/vespa/modelintegration/evaluator/OnnxEvaluatorOptions.java
+++ b/model-integration/src/main/java/ai/vespa/modelintegration/evaluator/OnnxEvaluatorOptions.java
@@ -41,6 +41,7 @@ public class OnnxEvaluatorOptions {
         options.setExecutionMode(executionMode);
         options.setInterOpNumThreads(executionMode == PARALLEL ? interOpThreads : 1);
         options.setIntraOpNumThreads(intraOpThreads);
+        options.setCPUArenaAllocator(false);
         if (loadCuda) {
             options.addCUDA(gpuDeviceNumber);
         }