[CI] Add test for paddleocr_vl (#4627)

fastdeploy/input/paddleocr_vl_processor/process.py

@@ -147,6 +147,7 @@ class DataProcessor:
             "num_input_video_tokens": 0,
             "fps": [],
             "mm_positions": [],
+            "mm_hashes": [],
             "vit_seqlen": [],
             "vit_position_ids": [],
         }

tests/e2e/test_paddleocr_vl_serving.py

@@ -0,0 +1,280 @@
+# 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 json
+import os
+import shutil
+import signal
+import socket
+import subprocess
+import sys
+import time
+
+import pytest
+import requests
+
+# Read ports from environment variables; use default values if not set
+FD_API_PORT = int(os.getenv("FD_API_PORT", 8188))
+FD_ENGINE_QUEUE_PORT = int(os.getenv("FD_ENGINE_QUEUE_PORT", 8133))
+FD_METRICS_PORT = int(os.getenv("FD_METRICS_PORT", 8233))
+FD_CACHE_QUEUE_PORT = int(os.getenv("FD_CACHE_QUEUE_PORT", 8333))
+
+# List of ports to clean before and after tests
+PORTS_TO_CLEAN = [FD_API_PORT, FD_ENGINE_QUEUE_PORT, FD_METRICS_PORT, FD_CACHE_QUEUE_PORT]
+
+os.environ["FD_USE_MACHETE"] = "0"
+
+
+def is_port_open(host: str, port: int, timeout=1.0):
+    """
+    Check if a TCP port is open on the given host.
+    Returns True if connection succeeds, False otherwise.
+    """
+    try:
+        with socket.create_connection((host, port), timeout):
+            return True
+    except Exception:
+        return False
+
+
+def kill_process_on_port(port: int):
+    """
+    Kill processes that are listening on the given port.
+    Uses `lsof` to find process ids and sends SIGKILL.
+    """
+    try:
+        output = subprocess.check_output(f"lsof -i:{port} -t", shell=True).decode().strip()
+        current_pid = os.getpid()
+        parent_pid = os.getppid()
+        for pid in output.splitlines():
+            pid = int(pid)
+            if pid in (current_pid, parent_pid):
+                print(f"Skip killing current process (pid={pid}) on port {port}")
+                continue
+            os.kill(pid, signal.SIGKILL)
+            print(f"Killed process on port {port}, pid={pid}")
+    except subprocess.CalledProcessError:
+        pass
+
+
+def clean_ports():
+    """
+    Kill all processes occupying the ports listed in PORTS_TO_CLEAN.
+    """
+    for port in PORTS_TO_CLEAN:
+        kill_process_on_port(port)
+    time.sleep(2)
+
+
+@pytest.fixture(scope="session", autouse=True)
+def setup_and_run_server():
+    """
+    Pytest fixture that runs once per test session:
+    - Cleans ports before tests
+    - Starts the API server as a subprocess
+    - Waits for server port to open (up to 30 seconds)
+    - Tears down server after all tests finish
+    """
+    print("Pre-test port cleanup...")
+    clean_ports()
+    print("log dir clean ")
+    if os.path.exists("log") and os.path.isdir("log"):
+        shutil.rmtree("log")
+
+    base_path = os.getenv("MODEL_PATH")
+    if base_path:
+        model_path = os.path.join(base_path, "PaddleOCR-VL-0.9B")
+    else:
+        model_path = "./PaddleOCR-VL-0.9B"
+
+    log_path = "server.log"
+    limit_mm_str = json.dumps({"image": 100, "video": 100})
+
+    cmd = [
+        sys.executable,
+        "-m",
+        "fastdeploy.entrypoints.openai.api_server",
+        "--model",
+        model_path,
+        "--port",
+        str(FD_API_PORT),
+        "--tensor-parallel-size",
+        "2",
+        "--engine-worker-queue-port",
+        str(FD_ENGINE_QUEUE_PORT),
+        "--metrics-port",
+        str(FD_METRICS_PORT),
+        "--cache-queue-port",
+        str(FD_CACHE_QUEUE_PORT),
+        "--enable-mm",
+        "--max-model-len",
+        "32768",
+        "--max-num-batched-tokens",
+        "384",
+        "--max-num-seqs",
+        "128",
+        "--limit-mm-per-prompt",
+        limit_mm_str,
+        "--enable-chunked-prefill",
+        "--kv-cache-ratio",
+        "0.71",
+        "--quantization",
+        "wint4",
+        "--graph-optimization-config",
+        '{"graph_opt_level":0, "use_cudagraph":true}',
+    ]
+
+    # Start subprocess in new process group
+    with open(log_path, "w") as logfile:
+        process = subprocess.Popen(
+            cmd,
+            stdout=logfile,
+            stderr=subprocess.STDOUT,
+            start_new_session=True,  # Enables killing full group via os.killpg
+        )
+
+    # Wait up to 10 minutes for API server to be ready
+    for _ in range(10 * 60):
+        if is_port_open("127.0.0.1", FD_API_PORT):
+            print(f"API server is up on port {FD_API_PORT}")
+            break
+        time.sleep(1)
+    else:
+        print("[TIMEOUT] API server failed to start in 5 minutes. Cleaning up...")
+        try:
+            os.killpg(process.pid, signal.SIGTERM)
+        except Exception as e:
+            print(f"Failed to kill process group: {e}")
+        raise RuntimeError(f"API server did not start on port {FD_API_PORT}")
+
+    yield  # Run tests
+
+    print("\n===== Post-test server cleanup... =====")
+    try:
+        os.killpg(process.pid, signal.SIGTERM)
+        print(f"API server (pid={process.pid}) terminated")
+        clean_ports()
+    except Exception as e:
+        print(f"Failed to terminate API server: {e}")
+
+
+@pytest.fixture(scope="session")
+def api_url(request):
+    """
+    Returns the API endpoint URL for chat completions.
+    """
+    return f"http://0.0.0.0:{FD_API_PORT}/v1/chat/completions"
+
+
+@pytest.fixture(scope="session")
+def metrics_url(request):
+    """
+    Returns the metrics endpoint URL.
+    """
+    return f"http://0.0.0.0:{FD_METRICS_PORT}/metrics"
+
+
+@pytest.fixture
+def headers():
+    """
+    Returns common HTTP request headers.
+    """
+    return {"Content-Type": "application/json"}
+
+
+@pytest.fixture
+def consistent_payload():
+    """
+    Returns a fixed payload for consistency testing,
+    including a fixed random seed and temperature.
+    """
+    return {
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "image_url",
+                        "image_url": {
+                            "url": "https://paddle-model-ecology.bj.bcebos.com/PPOCRVL/dataset/ocr_v5_eval/handwrite_ch_rec_val/中文手写古籍_000054_crop_32.jpg",
+                        },
+                    },
+                    {"type": "text", "text": "OCR:"},
+                ],
+            }
+        ],
+        "temperature": 0.8,
+        "top_p": 0,  # fix top_p to reduce randomness
+        "seed": 13,  # fixed random seed
+    }
+
+
+# ==========================
+# Helper function to calculate difference rate between two texts
+# ==========================
+def calculate_diff_rate(text1, text2):
+    """
+    Calculate the difference rate between two strings
+    based on the normalized Levenshtein edit distance.
+    Returns a float in [0,1], where 0 means identical.
+    """
+    if text1 == text2:
+        return 0.0
+
+    len1, len2 = len(text1), len(text2)
+    dp = [[0] * (len2 + 1) for _ in range(len1 + 1)]
+
+    for i in range(len1 + 1):
+        for j in range(len2 + 1):
+            if i == 0 or j == 0:
+                dp[i][j] = i + j
+            elif text1[i - 1] == text2[j - 1]:
+                dp[i][j] = dp[i - 1][j - 1]
+            else:
+                dp[i][j] = 1 + min(dp[i - 1][j], dp[i][j - 1], dp[i - 1][j - 1])
+
+    edit_distance = dp[len1][len2]
+    max_len = max(len1, len2)
+    return edit_distance / max_len if max_len > 0 else 0.0
+
+
+# ==========================
+# Consistency test for repeated runs with fixed payload
+# ==========================
+def test_consistency_between_runs(api_url, headers, consistent_payload):
+    """
+    Test that two runs with the same fixed input produce similar outputs.
+    """
+    # First request
+    resp1 = requests.post(api_url, headers=headers, json=consistent_payload)
+    assert resp1.status_code == 200
+    result1 = resp1.json()
+    content1 = result1["choices"][0]["message"]["content"]
+    print(content1)
+
+    # Second request
+    resp2 = requests.post(api_url, headers=headers, json=consistent_payload)
+    assert resp2.status_code == 200
+    result2 = resp2.json()
+
+    content2 = result2["choices"][0]["message"]["content"]
+    print(content2)
+
+    # Calculate difference rate
+    diff_rate = calculate_diff_rate(content1, content2)
+
+    # Verify that the difference rate is below the threshold
+    assert diff_rate < 0.05, f"Output difference too large ({diff_rate:.4%})"
+
+    assert content1 == "生甘草"