[Core][V0] Enable regex support with xgrammar

tests/entrypoints/llm/test_guided_generate.py

@@ -286,15 +286,26 @@ def test_validation_against_both_guided_decoding_options(sample_regex, llm):
 
 @pytest.mark.skip_global_cleanup
 def test_disable_guided_decoding_fallback(sample_regex, llm):
+    # see has_xgrammar_unsupported_json_features()
+    unsupported_json = {
+        "type": "object",
+        "properties": {
+            "example": {
+                "type": "string",
+                "minLength": 5  # unsupported by xgrammar
+            }
+        }
+    }
     sampling_params = SamplingParams(temperature=0.8,
                                      top_p=0.95,
                                      guided_decoding=GuidedDecodingParams(
-                                         regex=sample_regex,
+                                         json=unsupported_json,
                                          backend="xgrammar:no-fallback"))
 
     with pytest.raises(
             ValueError,
-            match="xgrammar does not support regex guided decoding"):
+            match="xgrammar does not support advanced JSON schema features "
+            "like enums, patterns or numeric ranges."):
         llm.generate(prompts="This should fail",
                      sampling_params=sampling_params,
                      use_tqdm=True)

vllm/model_executor/guided_decoding/__init__.py

@@ -59,14 +59,9 @@ def fallback_or_error(guided_params: GuidedDecodingParams, message: str,
         from vllm.model_executor.guided_decoding.xgrammar_decoding import (
             xgr_installed)
 
-        # xgrammar doesn't support regex, fallback to outlines
-        if guided_params.regex is not None:
-            fallback_or_error(
-                guided_params,
-                "xgrammar does not support regex guided decoding.", "outlines")
         # xgrammar doesn't support some JSON schema features
-        elif (guided_params.json is not None
-              and has_xgrammar_unsupported_json_features(guided_params.json)):
+        if (guided_params.json is not None and
+                has_xgrammar_unsupported_json_features(guided_params.json)):
             fallback_or_error(
                 guided_params,
                 "xgrammar does not support advanced JSON schema features like "

vllm/model_executor/guided_decoding/xgrammar_decoding.py

@@ -152,6 +152,7 @@ class GrammarConfig:
     grammar_str: str | None = None
     json_object: bool | None = None
     any_whitespace: bool = True
+    regex_str: str | None = None
     max_threads: int = 8
 
     @classmethod
@@ -255,6 +256,13 @@ def from_guided_params(cls,
                 max_threads=max_threads,
                 tokenizer_data=tokenizer_data,
             )
+        elif guided_params.regex:
+            return cls(
+                regex_str=guided_params.regex,
+                tokenizer_hash=tokenizer_hash,
+                max_threads=max_threads,
+                tokenizer_data=tokenizer_data,
+            )
         else:
             raise ValueError(
                 "Currently only support JSON and EBNF grammar mode for xgrammar"
@@ -330,6 +338,8 @@ def _ensure_ctx(self):
                 self.ctx = compiler\
                     .compile_json_schema('{"type": "object"}',
                                          any_whitespace=any_whitespace)
+            elif self.config.regex_str:
+                self.ctx = compiler.compile_regex(self.config.regex_str)
             else:
                 raise ValueError(
                     "Invalid configuration for xgrammar logits processor")