[Iluvatar][CI] Replace ci in ernie-300B-4layer with ernie-21b. (#6747)

scripts/run_ci_iluvatar.sh

@@ -5,7 +5,7 @@ echo "$DIR"
 ixsmi
 
 #先kill一遍
-ps -efww | grep -E 'run_ernie300B_4layer' | grep -v grep | awk '{print $2}' | xargs kill -9 || true
+ps -efww | grep -E 'run_ernie_21b' | grep -v grep | awk '{print $2}' | xargs kill -9 || true
 
 unset http_proxy
 unset https_proxy
@@ -41,7 +41,7 @@ echo "Run paddle.utils.run_check()"
 python -c "import paddle; paddle.utils.run_check()"
 
 INCLUDE_FOLDERS=(
-    "ERNIE_300B_4L"
+    # "ERNIE_300B_4L"
     "ERNIE-4.5-21B-A3B-Paddle"
     "ERNIE-4.5-VL-28B-A3B-Paddle"
     "PaddleOCR-VL"
@@ -59,15 +59,22 @@ for filename in "${INCLUDE_FOLDERS[@]}"; do
     cp -r $file $MODEL_DIR
 done
 
+CONTAINER_PP_DOC_DIR=/root/.paddlex/official_models
+mkdir -p $CONTAINER_PP_DOC_DIR
+echo "start copy $SOURCE_DIR/PP-DocLayoutV2 into $CONTAINER_PP_DOC_DIR"
+cp -r $SOURCE_DIR/PP-DocLayoutV2 $CONTAINER_PP_DOC_DIR
+
 echo "copy done"
 echo "ls $MODEL_DIR"
 ls $MODEL_DIR
+echo "ls $CONTAINER_PP_DOC_DIR"
+ls $CONTAINER_PP_DOC_DIR
 
 echo "build whl"
 bash build.sh || exit 1
 
 function print_error_message() {
-    if [ -f "log/launch_worker.0" ]; then
+    if [ -f "log/launch_worker.log" ]; then
         echo "------------------- log/launch_worker.log -----------------"
         cat log/launch_worker.log
     fi
@@ -92,7 +99,7 @@ export FD_SAMPLING_CLASS=rejection
 ################# Test offline ###################
 
 offline_ci_list=(
-    ${CI_PATH}/run_ernie300B_4layer.py
+    ${CI_PATH}/run_ernie_21b.py
     ${CI_PATH}/run_ernie_vl_28B.py
 )
 echo "test offline ci files: ${offline_ci_list[@]}"

tests/ci_use/iluvatar_UT/run_ernie_21b.py

@@ -21,20 +21,20 @@ from fastdeploy.utils import set_random_seed
 tests_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
 sys.path.insert(0, tests_dir)
 
-from ci_use.iluvatar_UT.utils import TIMEOUT_MSG, timeout
+from ci_use.iluvatar_UT.utils import timeout
 
 
-@timeout(80)
+@timeout(240)
 def offline_infer_check():
     set_random_seed(123)
 
     prompts = [
-        "Hello, my name is",
+        "The largest ocean is",
     ]
-    sampling_params = SamplingParams(temperature=0.8, top_p=0.00001, max_tokens=16)
+    sampling_params = SamplingParams(temperature=0.8, top_p=0.00001, max_tokens=128)
     graph_optimization_config = {"use_cudagraph": False}
     llm = LLM(
-        model="/model_data/ERNIE_300B_4L",
+        model="/model_data/ERNIE-4.5-21B-A3B-Paddle",
         tensor_parallel_size=2,
         max_model_len=8192,
         quantization="wint8",
@@ -43,24 +43,11 @@ def offline_infer_check():
     )
     outputs = llm.generate(prompts, sampling_params)
 
-    assert outputs[0].outputs.token_ids == [
+    for output in outputs:
-        23768,
+        generated_text = output.outputs.text
-        97000,
+        print(f"generated_text={generated_text}")
-        47814,
+        assert "pacific ocean" in generated_text.lower()
-        59335,
+
-        68170,
-        183,
-        97404,
-        100088,
-        36310,
-        95633,
-        95913,
-        41459,
-        95049,
-        94970,
-        96840,
-        2,
-    ], f"{outputs[0].outputs.token_ids}"
     print("PASSED")
 
 
@@ -69,7 +56,6 @@ if __name__ == "__main__":
         result = offline_infer_check()
         sys.exit(0)
     except TimeoutError:
-        print(TIMEOUT_MSG)
         sys.exit(124)
     except Exception:
         sys.exit(1)

tests/ci_use/iluvatar_UT/run_ernie_vl_28B.py

@@ -26,7 +26,7 @@ from fastdeploy.utils import set_random_seed
 tests_dir = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", ".."))
 sys.path.insert(0, tests_dir)
 
-from ci_use.iluvatar_UT.utils import TIMEOUT_MSG, timeout
+from ci_use.iluvatar_UT.utils import timeout
 
 
 @timeout(240)
@@ -97,7 +97,6 @@ if __name__ == "__main__":
         result = offline_infer_check()
         sys.exit(0)
     except TimeoutError:
-        print(TIMEOUT_MSG)
         sys.exit(124)
     except Exception:
         sys.exit(1)

tests/ci_use/iluvatar_UT/utils.py

@@ -23,6 +23,3 @@ def timeout(seconds):
         return wrapper
 
     return decorator
-
-
-TIMEOUT_MSG = "The timeout exit may be due to multiple processes sharing the same gpu card. You can check this using ixsmi on the device."