Clean up detokenizer and remove dead multimodal_gen code (#21588)

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

test/registered/bench_fn/test_bench_serving_functionality.py

@@ -8,6 +8,7 @@ from pathlib import Path
 
 from sglang.bench_serving import run_benchmark
 from sglang.benchmark.utils import parse_custom_headers
+from sglang.srt.constants import HEALTH_CHECK_RID_PREFIX
 from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
@@ -80,7 +81,7 @@ class TestBenchServingFunctionality(CustomTestCase):
                 continue
             text = obj.get("obj", {}).get("text")
             rid = obj.get("rid", "")
-            if text and not rid.startswith("HEALTH_CHECK"):
+            if text and not rid.startswith(HEALTH_CHECK_RID_PREFIX):
                 reqs.append(text)
 
         self.assertGreaterEqual(len(reqs), NUM_CONVERSATIONS * NUM_TURNS)

test/registered/unit/observability/test_request_metrics_exporter.py

@@ -58,6 +58,7 @@ import tempfile
 import unittest
 from unittest.mock import MagicMock, patch
 
+from sglang.srt.constants import HEALTH_CHECK_RID_PREFIX
 from sglang.srt.observability.request_metrics_exporter import (
     FileRequestMetricsExporter,
     RequestMetricsExporter,
@@ -243,7 +244,7 @@ class TestFileRequestMetricsExporter(unittest.TestCase):
 
     def test_write_record_skips_health_check(self):
         exporter = self._make_exporter()
-        obj = _GenerateReqInput(rid="HEALTH_CHECK_123", text="ping")
+        obj = _GenerateReqInput(rid=f"{HEALTH_CHECK_RID_PREFIX}_123", text="ping")
         asyncio.run(exporter.write_record(obj, {}))
 
         files = os.listdir(self.tmp_dir)

test/registered/utils/test_request_logger.py

@@ -8,6 +8,7 @@ from pathlib import Path
 
 import requests
 
+from sglang.srt.constants import HEALTH_CHECK_RID_PREFIX
 from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.test_utils import (
@@ -197,7 +198,7 @@ class TestRequestLoggerJson(BaseTestRequestLogger, CustomTestCase):
                 continue
 
             rid = data.get("rid", "")
-            if rid.startswith("HEALTH_CHECK"):
+            if rid.startswith(HEALTH_CHECK_RID_PREFIX):
                 continue
 
             if data.get("event") == "request.received":