Deprecate V0 engine

Author: yinggeh

ci/L0_backend_vllm/accuracy_test/test.sh

@@ -48,17 +48,11 @@ RET=0
 set +e
 # Need to generate baseline first, since running 2 vLLM engines causes
 # memory issues: https://github.com/vllm-project/vllm/issues/2248
-export VLLM_USE_V1=0
-export VLLM_WORKER_MULTIPROC_METHOD=spawn
 python3 $CLIENT_PY --generate-baseline >> $VLLM_ENGINE_LOG 2>&1 & BASELINE_PID=$!
 wait $BASELINE_PID
 
 python3 $CLIENT_PY --generate-guided-baseline > $VLLM_ENGINE_LOG 2>&1 & BASELINE_PID=$!
 wait $BASELINE_PID
-
-unset VLLM_USE_V1
-unset VLLM_WORKER_MULTIPROC_METHOD
-
 set -e
 
 run_server
@@ -88,12 +82,6 @@ set -e
 kill $SERVER_PID
 wait $SERVER_PID
 
-# Check that warning about V1 Engine appears in log - this warning is expected
-if ! grep -q "Engine in background thread is experimental on VLLM_USE_V1=1. Falling back to V0 Engine." $SERVER_LOG; then
-    echo -e "\n***\n*** ERROR: Expected warning about vLLM falling back to V0 Engine not found in logs.\n***"
-    RET=1
-fi
-
 rm -rf models/
 
 if [ $RET -eq 1 ]; then

ci/L0_check_health_vllm/test.sh

@@ -48,23 +48,24 @@ function enable_health_check {
 }
 
 VLLM_INSTALL_PATH="/usr/local/lib/python3.12/dist-packages/vllm"
+VLLM_V1_ENGINE_PATH="$VLLM_INSTALL_PATH/v1/engine"
 
 function mock_vllm_async_llm_engine {
     # backup original file
-    mv $VLLM_INSTALL_PATH/engine/multiprocessing/client.py $VLLM_INSTALL_PATH/engine/multiprocessing/client.py.backup
-    cp $VLLM_INSTALL_PATH/engine/multiprocessing/client.py.backup $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
+    mv $VLLM_V1_ENGINE_PATH/async_llm.py $VLLM_V1_ENGINE_PATH/async_llm.py.backup
+    cp $VLLM_V1_ENGINE_PATH/async_llm.py.backup $VLLM_V1_ENGINE_PATH/async_llm.py
     # overwrite the original check_health method
-    echo -e "" >> $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
-    echo -e "    async def check_health(self, check_count=[0]):" >> $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
-    echo -e "        check_count[0] += 1" >> $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
-    echo -e "        if check_count[0] > 1:" >> $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
-    echo -e "            raise RuntimeError(\"Simulated vLLM check_health() failure\")" >> $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
+    echo -e "" >> $VLLM_V1_ENGINE_PATH/async_llm.py
+    echo -e "    async def check_health(self, check_count=[0]):" >> $VLLM_V1_ENGINE_PATH/async_llm.py
+    echo -e "        check_count[0] += 1" >> $VLLM_V1_ENGINE_PATH/async_llm.py
+    echo -e "        if check_count[0] > 1:" >> $VLLM_V1_ENGINE_PATH/async_llm.py
+    echo -e "            raise RuntimeError(\"Simulated vLLM check_health() failure\")" >> $VLLM_V1_ENGINE_PATH/async_llm.py
 }
 
 function unmock_vllm_async_llm_engine {
     # restore from backup
-    rm -f $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
-    mv $VLLM_INSTALL_PATH/engine/multiprocessing/client.py.backup $VLLM_INSTALL_PATH/engine/multiprocessing/client.py
+    rm -f $VLLM_V1_ENGINE_PATH/async_llm.py
+    mv $VLLM_V1_ENGINE_PATH/async_llm.py.backup $VLLM_V1_ENGINE_PATH/async_llm.py
 }
 
 function test_check_health {