From a5149b2feeb74c15990dafb1d8bf16cdb98616e4 Mon Sep 17 00:00:00 2001
From: "Kevin H. Luu" <khluu000@gmail.com>
Date: Tue, 4 Aug 2026 14:13:59 -0700
Subject: [PATCH] [CI] Stabilize GLM-5.2 PCP evaluation (#51015)

Signed-off-by: khluu <khluu000@gmail.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
---
 tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP1-PCP4-EP.yaml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP1-PCP4-EP.yaml b/tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP1-PCP4-EP.yaml
index 14b4f2c0d1..d4ddfcc99f 100644
--- a/tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP1-PCP4-EP.yaml
+++ b/tests/evals/gsm8k/configs/GLM-5.2-NVFP4-TP1-PCP4-EP.yaml
@@ -13,5 +13,6 @@ server_args: >-
   --enable-expert-parallel
   --kv-cache-dtype fp8
 env:
+  PYTORCH_CUDA_ALLOC_CONF: "expandable_segments:True"
   VLLM_LOGGING_LEVEL: "DEBUG"
   VLLM_USE_V2_MODEL_RUNNER: "1"