[GPT-OSS] Structure_Tag support for gpt-oss tool-call in cot (#25515)

Signed-off-by: Hanchenli <lihanc2002@gmail.com>
Signed-off-by: Hanchenli <61769611+Hanchenli@users.noreply.github.com>
Signed-off-by: Wei Wei <wwei6@meta.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Wei Wei <wwei6@meta.com>
Co-authored-by: Wei Wei <weiweinpu@gmail.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>

tests/v1/entrypoints/llm/test_struct_output_generate.py

@@ -864,3 +864,49 @@ def test_structured_output_batched_with_non_structured_outputs_requests(
             # non-structured outputs requests should not return a valid JSON here
             with pytest.raises(ValueError):
                 output_json = json.loads(generated_text)
+
+
+@pytest.mark.parametrize("guided_decoding_backend", ["xgrammar"])
+def test_structured_output_with_structural_tag(
+    monkeypatch: pytest.MonkeyPatch,
+    guided_decoding_backend: str,
+):
+    monkeypatch.setenv("VLLM_USE_V1", "1")
+
+    llm = LLM(
+        model="Qwen/Qwen2.5-1.5B-Instruct",
+        guided_decoding_backend=guided_decoding_backend,
+    )
+
+    structural_tag_config = {
+        "type": "structural_tag",
+        "format": {
+            "type": "triggered_tags",
+            "tags": [
+                {"begin": "hello_flag", "content": {"type": "any_text"}, "end": "hello"}
+            ],
+            "triggers": ["hello"],
+            "stop_after_first": False,
+        },
+    }
+
+    sampling_params = SamplingParams(
+        temperature=0.0,
+        max_tokens=500,
+        guided_decoding=StructuredOutputsParams(
+            structural_tag=json.dumps(structural_tag_config)
+        ),
+    )
+
+    prompt = "Hello and repete hello 10 times, do not say anything else. Only say hello hello hello, now start"
+    outputs = llm.generate(prompt, sampling_params=sampling_params, use_tqdm=True)
+    assert outputs is not None
+    for output in outputs:
+        assert output is not None
+        assert isinstance(output, RequestOutput)
+        prompt = output.prompt
+        generated_text = output.outputs[0].text
+        assert generated_text is not None
+        assert "hello_flag" in generated_text, (
+            f"Expected 'hello_flag' to be in generated text, but got: {generated_text}"
+        )