From 4a8ae26e532c4b36acc1dd2f5e7c672d73832b4e Mon Sep 17 00:00:00 2001
From: Andreas Karatzas <akaratza@amd.com>
Date: Tue, 5 May 2026 20:08:12 -0500
Subject: [PATCH] [ROCm][CI] Use vLLM generation defaults for DeepSeek
 prefetch-offload eval (#41575)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>
---
 .../deepseek_v2_lite_prefetch_offload.sh                         | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh b/.buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh
index de48eb282a..0eadfa1f80 100755
--- a/.buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh
+++ b/.buildkite/scripts/scheduled_integration_test/deepseek_v2_lite_prefetch_offload.sh
@@ -51,6 +51,7 @@ vllm serve "$MODEL" \
   --offload-num-in-group 2 \
   --offload-prefetch-step 1 \
   --offload-params w13_weight w2_weight \
+  --generation-config vllm \
   --port "$PORT" \
   ${EXTRA_ARGS+"${EXTRA_ARGS[@]}"} &
 SERVER_PID=$!