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

Signed-off-by: Crucifixion-Fxl <[email protected]>
Co-authored-by: Crucifixion-Fxl <[email protected]>

Author: Crucifixion-Fxl

.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 @@ def process_one_metric(
 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 _eval_correctness_choice(expected, actual):
         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 @@
 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

@@ -1,3 +1,4 @@
+regex # Replace re for higher-performance regex matching
 cachetools
 psutil
 sentencepiece  # Required for LLaMA tokenizer.

requirements/nightly_torch_test.txt

@@ -38,4 +38,4 @@ matplotlib # required for qwen-vl test
 # required for  Multi-Modal Models Test (Standard)
 num2words # required for smolvlm test
 pqdm
-timm # required for internvl test
+timm # required for internvl test

setup.py

@@ -5,12 +5,12 @@
 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 @@ def run(self) -> None:
             # 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

@@ -1,12 +1,12 @@
 # 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

@@ -1,8 +1,6 @@
 # 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 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

@@ -1,10 +1,9 @@
 # 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,
-                           match=re.compile(
-                               '.*Guided decoding .* multi-step decoding.*')):
+        with pytest.raises(
+                openai.BadRequestError,
+                match=re.compile(
+                    '.*Guided decoding .* multi-step decoding.*').pattern):
             await client.completions.create(
                 model=model_name,
                 prompt="Hello",

tests/models/multimodal/generation/test_phi4mm.py

@@ -1,12 +1,12 @@
 # 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

@@ -1,10 +1,10 @@
 # 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,
@@ -333,4 +333,4 @@ def test_streaming_output_valid(output, empty_params, delta_len):
             combined_messages += message.tool_calls[0].function.arguments
     combined_messages += "}]"
     assert json.loads(combined_messages) == output
-    assert json.dumps(json.loads(combined_messages)) == output_json
+    assert json.dumps(json.loads(combined_messages)) == output_json

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

@@ -1,11 +1,11 @@
 # 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

@@ -1,9 +1,10 @@
 # SPDX-License-Identifier: Apache-2.0
 
-import re
 from enum import Enum
 from typing import Optional
 
+import regex as re
+
 from vllm import CompletionOutput
 
 

vllm/collect_env.py

@@ -815,4 +815,4 @@ def main():
 
 
 if __name__ == '__main__':
-    main()
+    main()

vllm/config.py

@@ -6,7 +6,6 @@
 import hashlib
 import inspect
 import json
-import re
 import textwrap
 import uuid
 import warnings
@@ -20,6 +19,7 @@
 from typing import (TYPE_CHECKING, Any, Callable, ClassVar, Literal, Optional,
                     Protocol, TypeVar, Union, cast, get_args, get_origin)
 
+import regex as re
 import torch
 from torch.distributed import ProcessGroup, ReduceOp
 from transformers import PretrainedConfig

vllm/engine/arg_utils.py

@@ -4,7 +4,6 @@
 import argparse
 import dataclasses
 import json
-import re
 import sys
 import threading
 import warnings
@@ -13,6 +12,7 @@
 from typing import (Annotated, Any, Callable, Dict, List, Literal, Optional,
                     Type, TypeVar, Union, cast, get_args, get_origin)
 
+import regex as re
 import torch
 from typing_extensions import TypeIs, deprecated
 

vllm/entrypoints/openai/api_server.py

@@ -7,7 +7,6 @@
 import inspect
 import multiprocessing
 import os
-import re
 import signal
 import socket
 import tempfile
@@ -21,6 +20,7 @@
 from typing import Annotated, Optional, Union
 
 import prometheus_client
+import regex as re
 import uvloop
 from fastapi import APIRouter, Depends, FastAPI, Form, HTTPException, Request
 from fastapi.exceptions import RequestValidationError

vllm/entrypoints/openai/protocol.py

@@ -3,11 +3,11 @@
 # Adapted from
 # https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
 import json
-import re
 import time
 from http import HTTPStatus
 from typing import Annotated, Any, ClassVar, Literal, Optional, Union
 
+import regex as re
 import torch
 from fastapi import HTTPException, UploadFile
 from pydantic import (BaseModel, ConfigDict, Field, TypeAdapter,

vllm/entrypoints/openai/serving_chat.py

@@ -2,14 +2,14 @@
 
 import asyncio
 import json
-import re
 import time
 from collections.abc import AsyncGenerator, AsyncIterator
 from collections.abc import Sequence as GenericSequence
 from typing import Callable, Final, Optional, Union
 
 import jinja2
 import partial_json_parser
+import regex as re
 from fastapi import Request
 from pydantic import TypeAdapter
 

vllm/entrypoints/openai/tool_parsers/deepseekv3_tool_parser.py

@@ -1,9 +1,10 @@
 # SPDX-License-Identifier: Apache-2.0
 
-import re
 from collections.abc import Sequence
 from typing import Union
 
+import regex as re
+
 from vllm.entrypoints.openai.protocol import (ChatCompletionRequest,
                                               DeltaFunctionCall, DeltaMessage,
                                               DeltaToolCall,

vllm/entrypoints/openai/tool_parsers/granite_20b_fc_tool_parser.py

@@ -1,12 +1,12 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import json
-import re
 from collections.abc import Sequence
 from json import JSONDecoder
 from typing import Union
 
 import partial_json_parser
+import regex as re
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import random_tool_call_id

vllm/entrypoints/openai/tool_parsers/hermes_tool_parser.py

@@ -1,11 +1,11 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import json
-import re
 from collections.abc import Sequence
 from typing import Union
 
 import partial_json_parser
+import regex as re
 from partial_json_parser.core.options import Allow
 
 from vllm.entrypoints.chat_utils import random_tool_call_id