[Bugfix] Migrate to REGEX Library to prevent catastrophic backtracking (#18454)

Signed-off-by: Crucifixion-Fxl <xmufxl@gmail.com>
Co-authored-by: Crucifixion-Fxl <xmufxl@gmail.com>

.github/scripts/cleanup_pr_body.sh

@@ -26,7 +26,7 @@ sed -i '/\*\*BEFORE SUBMITTING, PLEASE READ.*\*\*/,$d' "${NEW}"
 
 # Remove HTML <details> section that includes <summary> text of "PR Checklist (Click to Expand)"
 python3 - <<EOF
-import re
+import regex as re
 
 with open("${NEW}", "r") as file:
     content = file.read()

benchmarks/benchmark_serving_structured_output.py

@@ -672,7 +672,7 @@ async def benchmark(
 def evaluate(ret, args):
     def _eval_correctness_json(expected, actual):
         # extract json string from string using regex
-        import re
+        import regex as re
 
         actual = actual.replace("\n", "").replace(" ", "").strip()
         try:
@@ -687,7 +687,7 @@ def evaluate(ret, args):
         return actual in args.choice
 
     def _eval_correctness_regex(expected, actual):
-        import re
+        import regex as re
 
         return re.match(args.regex, actual) is not None
 

benchmarks/kernels/graph_machete_bench.py

@@ -2,11 +2,11 @@
 
 import math
 import pickle
-import re
 from collections import defaultdict
 
 import matplotlib.pyplot as plt
 import pandas as pd
+import regex as re
 import seaborn as sns
 from torch.utils.benchmark import Measurement as TMeasurement
 

examples/offline_inference/prithvi_geospatial_mae.py

@@ -20,12 +20,12 @@ python prithvi_geospatial_mae.py
 import argparse
 import datetime
 import os
-import re
 from typing import Union
 
 import albumentations
 import numpy as np
 import rasterio
+import regex as re
 import torch
 from einops import rearrange
 from terratorch.datamodules import Sen1Floods11NonGeoDataModule

pyproject.toml

@@ -8,6 +8,7 @@ requires = [
     "setuptools-scm>=8.0",
     "torch == 2.7.0",
     "wheel",
+    "regex",
     "jinja2",
 ]
 build-backend = "setuptools.build_meta"

requirements/build.txt

@@ -7,3 +7,4 @@ setuptools-scm>=8
 torch==2.7.0
 wheel
 jinja2>=3.1.6
+regex

requirements/common.txt

+regex # Replace re for higher-performance regex matching
 cachetools
 psutil
 sentencepiece  # Required for LLaMA tokenizer.

setup.py

@@ -5,12 +5,12 @@ import importlib.util
 import json
 import logging
 import os
-import re
 import subprocess
 import sys
 from pathlib import Path
 from shutil import which
 
+import regex as re
 import torch
 from packaging.version import Version, parse
 from setuptools import Extension, setup
@@ -389,7 +389,6 @@ class repackage_wheel(build_ext):
             # vllm_flash_attn python code:
             # Regex from
             #  `glob.translate('vllm/vllm_flash_attn/**/*.py', recursive=True)`
-            import re
             compiled_regex = re.compile(
                 r"vllm/vllm_flash_attn/(?:[^/.][^/]*/)*(?!\.)[^/]*\.py")
             file_members += list(

tests/entrypoints/llm/test_guided_generate.py

 # SPDX-License-Identifier: Apache-2.0
 
 import json
-import re
 import weakref
 from enum import Enum
 
 import jsonschema
 import pytest
+import regex as re
 from pydantic import BaseModel
 
 from vllm.distributed import cleanup_dist_env_and_memory

tests/entrypoints/openai/test_chat.py

@@ -2,13 +2,13 @@
 
 # imports for guided decoding tests
 import json
-import re
 from typing import Optional
 
 import jsonschema
 import openai  # use the official client for correctness check
 import pytest
 import pytest_asyncio
+import regex as re
 import requests
 import torch
 from openai import BadRequestError, OpenAI

tests/entrypoints/openai/test_completion.py

 # SPDX-License-Identifier: Apache-2.0
-
 # imports for guided decoding tests
 import json
-import re
 import shutil
 from tempfile import TemporaryDirectory
 from typing import Optional
@@ -11,6 +9,7 @@ import jsonschema
 import openai  # use the official client for correctness check
 import pytest
 import pytest_asyncio
+import regex as re
 # downloading lora to test lora requests
 from huggingface_hub import snapshot_download
 from openai import BadRequestError

tests/entrypoints/openai/test_prompt_validation.py

 # SPDX-License-Identifier: Apache-2.0
 
 # imports for guided decoding tests
-import re
-
 import openai
 import pytest
+import regex as re
 
 from ...utils import RemoteOpenAIServer
 
@@ -32,7 +31,7 @@ async def test_out_of_vocab_token_ids():
         client = remote_server.get_async_client()
 
         with pytest.raises(openai.BadRequestError,
-                           match=re.compile('.*out of vocabulary.*')):
+                           match=re.compile('.*out of vocabulary.*').pattern):
             await client.completions.create(model=model_name,
                                             prompt=[999999],
                                             max_tokens=5,
@@ -46,9 +45,10 @@ async def test_reject_multistep_with_guided_decoding():
     with RemoteOpenAIServer(model_name, server_args) as remote_server:
         client = remote_server.get_async_client()
 
-        with pytest.raises(openai.BadRequestError,
+        with pytest.raises(
+                openai.BadRequestError,
                 match=re.compile(
-                               '.*Guided decoding .* multi-step decoding.*')):
+                    '.*Guided decoding .* multi-step decoding.*').pattern):
             await client.completions.create(
                 model=model_name,
                 prompt="Hello",

tests/models/multimodal/generation/test_phi4mm.py

 # SPDX-License-Identifier: Apache-2.0
 
 import os
-import re
 from collections.abc import Sequence
 from typing import Optional
 
 import librosa
 import pytest
+import regex as re
 from huggingface_hub import snapshot_download
 from transformers import AutoTokenizer
 

tests/models/multimodal/generation/vlm_utils/model_utils.py

@@ -3,11 +3,11 @@
 for manipulating the input / output of HF & vLLM test runners, which are
 typically specific to a small subset of models.
 """
-import re
 import types
 from pathlib import PosixPath
 from typing import Optional, Union
 
+import regex as re
 import torch
 from PIL.Image import Image
 from transformers import (AutoConfig, AutoTokenizer, BatchFeature,

tests/tool_use/test_tool_choice_required.py

 # SPDX-License-Identifier: Apache-2.0
 import json
-import re
 from copy import deepcopy
 from unittest.mock import MagicMock
 
 import pytest
+import regex as re
 from pydantic import TypeAdapter
 
 from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,

tests/v1/entrypoints/llm/test_struct_output_generate.py

@@ -4,12 +4,12 @@
 from __future__ import annotations
 
 import json
-import re
 from enum import Enum
 from typing import TYPE_CHECKING, Any
 
 import jsonschema
 import pytest
+import regex as re
 from pydantic import BaseModel
 
 from tests.reasoning.utils import run_reasoning_extraction

tests/v1/entrypoints/openai/test_completion.py

 # SPDX-License-Identifier: Apache-2.0
 
-import re
 from typing import Optional
 
 import openai  # use the official client for correctness check
 import pytest
 import pytest_asyncio
+import regex as re
 from openai import BadRequestError
 
 from tests.utils import RemoteOpenAIServer

tests/v1/sample/utils.py

 # SPDX-License-Identifier: Apache-2.0
 
-import re
 from enum import Enum
 from typing import Optional
 
+import regex as re
+
 from vllm import CompletionOutput