From 886c7632b212929896c733115b05f224958ea54f Mon Sep 17 00:00:00 2001
From: Xiangyu Tian <109123695+xiangyuT@users.noreply.github.com>
Date: Mon, 12 May 2025 13:44:33 +0800
Subject: [PATCH] Add IPEX_LLM_FORCE_BATCH_FORWARD for vLLM docker image
 (#13151)

---
 docker/llm/serving/xpu/docker/Dockerfile | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/docker/llm/serving/xpu/docker/Dockerfile b/docker/llm/serving/xpu/docker/Dockerfile
index e9ca4525..1dac2239 100644
--- a/docker/llm/serving/xpu/docker/Dockerfile
+++ b/docker/llm/serving/xpu/docker/Dockerfile
@@ -186,5 +186,8 @@ RUN set -eux && \
     # Re-enable sym_int4
     sed -i 's/qtype = ggml_tensor_qtype\["woq_int4"\]/qtype = ggml_tensor_qtype["sym_int4"]/' /usr/local/lib/python3.11/dist-packages/ipex_llm/transformers/convert.py
 
+# Set environment variables to force use batch_forward_new()
+ENV IPEX_LLM_FORCE_BATCH_FORWARD=1
+
 WORKDIR /llm/
 ENTRYPOINT ["bash", "/llm/start-vllm-service.sh"]