diff --git a/runs/frontier-s3-real-v0/run_full_real.sh b/runs/frontier-s3-real-v0/run_full_real.sh
index 1b20fa7..c4b31b2 100644
--- a/runs/frontier-s3-real-v0/run_full_real.sh
+++ b/runs/frontier-s3-real-v0/run_full_real.sh
@@ -13,7 +13,7 @@ SERVED_MODEL="qwen3-30b-s3-real-full"
 MAX_MODEL_LEN="${MAX_MODEL_LEN:-40960}"
 ALLOW_SYNTHETIC_PROMPTS="${ALLOW_SYNTHETIC_PROMPTS:-false}"
 ALLOW_LONG_CONTEXT_SERVER="${ALLOW_LONG_CONTEXT_SERVER:-false}"
-FLASHINFER_WORKSPACE_BASE="/tmp/wjh/flashinfer-frontier-s3-real-v0-${RHO}-${CONFIG}-t${TRIAL}"
+FLASHINFER_WORKSPACE_BASE="${FLASHINFER_WORKSPACE_BASE:-/tmp/wjh/flashinfer-frontier-s3-real-v0-${CONFIG}}"
 SERVER_PID=""
 TELEMETRY_PID=""
 
@@ -45,6 +45,9 @@ if (( MAX_MODEL_LEN > 40960 )) && [[ "${ALLOW_LONG_CONTEXT_SERVER}" != "true" ]]
   echo "ERROR: MAX_MODEL_LEN>40960 requires ALLOW_LONG_CONTEXT_SERVER=true" >&2
   exit 1
 fi
+if (( MAX_MODEL_LEN > 40960 )); then
+  export VLLM_CACHE_ROOT="${VLLM_CACHE_ROOT:-/tmp/wjh/vllm-cache-frontier-s3-real-longctx}"
+fi
 
 TRACE_INPUT_ROOT="${TRACE_INPUT_ROOT:-${CONTROL_ROOT}/inputs/${INPUT_NAME}}"
 REQUESTS_FILE="${TRACE_INPUT_ROOT}/real_requests.jsonl"