update ci cases

.github/workflows/ci.yml

@@ -60,10 +60,12 @@ jobs:
           runner_name="${{ runner.name }}"
           last_char="${runner_name: -1}"
 
-          if [[ "$last_char" =~ [0-3] ]]; then
+          if [ "${last_char}" = "1" ]; then
-            gpu_id="$last_char"
+            gpu_id=2
+            DEVICES="2,3"
           else
-            gpu_id="0"  
+            gpu_id=0
+            DEVICES="0,1"
           fi
           FD_API_PORT=$((9180 + gpu_id * 100))
           FD_ENGINE_QUEUE_PORT=$((9150 + gpu_id * 100))
@@ -80,7 +82,7 @@ jobs:
           -e "FD_API_PORT=${FD_API_PORT}" \
           -e "FD_ENGINE_QUEUE_PORT=${FD_ENGINE_QUEUE_PORT}" \
           -e "FD_METRICS_PORT=${FD_METRICS_PORT}" \
-          --gpus device=${gpu_id} ${docker_image} /bin/bash -c "
+          --gpus '"device=${DEVICES}"' ${docker_image} /bin/bash -c "
           git config --global --add safe.directory /workspace/FastDeploy
           cd FastDeploy
           bash scripts/run_ci.sh

scripts/run_ci.sh

@@ -5,7 +5,7 @@ echo "$DIR"
 python -m pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
 python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu126/
 python -m pip install -r requirements.txt
-python -m pip install jsonschema aistudio_sdk==0.2.6
+python -m pip install jsonschema aistudio_sdk==0.3.5
 bash build.sh || exit 1
 
 failed_files=()

test/ci_use/EB_VL_Lite/rollout_model.py

@@ -0,0 +1,69 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import difflib
+
+from paddleformers.trl.llm_utils import init_dist_env
+
+from fastdeploy.rl.rollout_config import RolloutModelConfig
+from fastdeploy.rl.rollout_model import RolloutModel
+
+_, ranks = init_dist_env()
+
+
+# base result
+base_path = os.getenv("MODEL_PATH")
+model_path = os.path.join(base_path, "ernie-4_5-vl-28b-a3b-bf16-paddle")
+
+# Usage example:
+init_kwargs = {
+    "model_name_or_path": MODEL_PATH,
+    "max_model_len": 32768,
+    "tensor_parallel_size": ranks,
+    "dynamic_load_weight": True,
+    "load_strategy": "ipc_snapshot",
+    "enable_mm": True,
+    "quantization": "wint8",
+}
+
+rollout_config = RolloutModelConfig(**init_kwargs)
+actor_eval_model = RolloutModel(rollout_config)
+
+content = ""
+for k, v in actor_eval_model.state_dict().items():
+    content += f"{k}\n"
+for k, v in actor_eval_model.get_name_mappings_to_training().items():
+    content += f"{k}:{v}\n"
+
+# with open("baseline.txt", "w", encoding="utf-8") as f:
+#     f.write(baseline)
+
+def compare_strings(a: str, b: str) -> bool:
+    if a == b:
+        print("✅ 两个字符串完全一致")
+        return True
+
+    print("❌ 字符串不一致，差异如下（上下文差异显示）：")
+    diff = difflib.ndiff(a.splitlines(), b.splitlines())
+    for line in diff:
+        if line.startswith("- ") or line.startswith("+ "):
+            print(line)
+
+    return False
+
+with open("baseline.txt", "r", encoding="utf-8") as f:
+    baseline = f.read()
+    assert compare_strings(baseline, content), "In the unittest of RL scenario, your modification " \
+        "caused inconsistency in the content before and after. Please fix it. " \
+        "Can request assistance from yuanlehome or gzy19990617 (github id)."

test/ci_use/EB_VL_Lite/test_EB_VL_Lite_serving.py

@@ -88,7 +88,7 @@ def setup_and_run_server():
         sys.executable, "-m", "fastdeploy.entrypoints.openai.api_server",
         "--model", model_path,
         "--port", str(FD_API_PORT),
-        "--tensor-parallel-size", "1",
+        "--tensor-parallel-size", "2",
         "--engine-worker-queue-port", str(FD_ENGINE_QUEUE_PORT),
         "--metrics-port", str(FD_METRICS_PORT),
         "--enable-mm",
@@ -197,9 +197,9 @@ def test_consistency_between_runs(api_url, headers, consistent_payload):
     # base result
     base_path = os.getenv("MODEL_PATH")
     if base_path:
-        base_file = os.path.join(base_path, "ernie-4_5-vl-base")
+        base_file = os.path.join(base_path, "ernie-4_5-vl-base-tp2")
     else:
-        base_file = "ernie-4_5-vl-base"
+        base_file = "ernie-4_5-vl-base-tp2"
     with open(base_file, "r") as f:
         content2 = f.read()
 

test/ci_use/EB_VL_Lite/test_rollout_model.py

@@ -0,0 +1,59 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import subprocess
+import sys
+import os
+import time
+import pytest
+
+
+def test_rollout_model_with_distributed_launch():
+    """
+    test_rollout_model
+    """
+    current_dir = os.path.dirname(os.path.abspath(__file__))
+
+    rollout_script = os.path.join(current_dir, "rollout_model.py")
+
+    command = [
+        sys.executable,
+        "-m", "paddle.distributed.launch",
+        "--gpus", "0,1",
+        rollout_script
+    ]
+
+    print(f"Executing command: {' '.join(command)}")
+
+    process = subprocess.Popen(
+        command,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
+        text=True
+    )
+
+    try:
+        stdout, stderr = process.communicate(timeout=300)
+        return_code = process.returncode
+    except subprocess.TimeoutExpired:
+        process.kill()
+        stdout, stderr = process.communicate()
+        return_code = -1
+
+    print("\n" + "=" * 50 + " STDOUT " + "=" * 50)
+    print(stdout)
+    print("\n" + "=" * 50 + " STDERR " + "=" * 50)
+    print(stderr)
+
+    assert return_code == 0, f"Process exited with code {return_code}"