[BugFix] Preserve tokenizer worker fanout when skip_tokenizer_init is enabled (#30682)

Co-authored-by: hjzhang <zhanghjzzz@qq.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>

test/registered/tokenizer/test_skip_tokenizer_init.py

@@ -36,7 +36,12 @@ class TestSkipTokenizerInit(CustomTestCase):
             cls.model,
             cls.base_url,
             timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=["--skip-tokenizer-init", "--incremental-streaming-output"],
+            other_args=[
+                "--skip-tokenizer-init",
+                "--incremental-streaming-output",
+                "--tokenizer-worker-num",
+                "4",
+            ],
         )
         cls.eos_token_id = [119690]
         cls.tokenizer = AutoTokenizer.from_pretrained(
@@ -216,7 +221,11 @@ class TestSkipTokenizerInitVLM(TestSkipTokenizerInit):
             cls.model,
             cls.base_url,
             timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=["--skip-tokenizer-init"],
+            other_args=[
+                "--skip-tokenizer-init",
+                "--tokenizer-worker-num",
+                "4",
+            ],
         )
         cls.eos_token_id = [cls.tokenizer.eos_token_id]