[XPU][CI]Change ci vl model to 28 b (#4764)

* Update XPU_VISIBLE_DEVICES and model parameters

* Update base response and adjust max tokens

* Implement process cleanup in CI workflow

Add process cleanup commands to prevent port conflicts

* Remove process cleanup commands from CI workflow

Removed old process cleanup commands to prevent port conflicts.

.github/workflows/ci_xpu.yml

@@ -68,7 +68,6 @@ jobs:
           FD_API_PORT=$((9180 + gpu_id * 100))
           FD_ENGINE_QUEUE_PORT=$((9150 + gpu_id * 100))
           FD_METRICS_PORT=$((9170 + gpu_id * 100))
-
           PARENT_DIR=$(dirname "$WORKSPACE")
           echo "PARENT_DIR:$PARENT_DIR"
           docker run --rm --net=host --cap-add=SYS_PTRACE --privileged --shm-size=64G  \

scripts/run_ci_xpu.sh

@@ -187,11 +187,11 @@ rm -f core*
 #清空消息队列
 ipcrm --all=msg
 echo "============================开始vl模型测试!============================"
-export XPU_VISIBLE_DEVICES="0,1,2,3,4,5,6,7"
+export XPU_VISIBLE_DEVICES="0,1,2,3"
 python -m fastdeploy.entrypoints.openai.api_server \
-    --model ${MODEL_PATH}/ERNIE-4.5-VL-424B-A47B-Paddle \
+    --model ${MODEL_PATH}/ERNIE-4.5-VL-28B-A3B-Paddle \
     --port 8188 \
-    --tensor-parallel-size 8 \
+    --tensor-parallel-size 4 \
     --max-model-len 32768 \
     --max-num-seqs 10 \
     --quantization wint8 \

tests/ci_use/XPU_45T/run_45vl.py

@@ -19,7 +19,7 @@ def test_45vl():
     ip = "0.0.0.0"
     service_http_port = "8188"  # 服务配置的
     client = openai.Client(base_url=f"http://{ip}:{service_http_port}/v1", api_key="EMPTY_API_KEY")
-    base_response = "\n这尊佛像是**北魏时期的释迦牟尼"
+    base_response = "\n\n这尊佛像属于**北魏时期**"
     # 非流式对话
     response = client.chat.completions.create(
         model="default",
@@ -39,7 +39,7 @@ def test_45vl():
         ],
         temperature=1,
         top_p=0,
-        max_tokens=75,
+        max_tokens=70,
         stream=False,
     )
     print(response.choices[0].message.content)