Revert "feat: reduce constrained-decoding overhead in TP" (#16845)

test/registered/constrained_decoding/test_constrained_decoding.py

@@ -5,8 +5,6 @@ register_amd_ci(est_time=179, suite="stage-b-test-small-1-gpu-amd")
 
 import unittest
 
-import torch
-
 from sglang.srt.utils import kill_process_tree
 from sglang.test.kits.ebnf_constrained_kit import TestEBNFConstrainedMinxin
 from sglang.test.kits.json_constrained_kit import TestJSONConstrainedMixin
@@ -23,7 +21,6 @@ from sglang.test.test_utils import (
 class ServerWithGrammar(CustomTestCase):
     backend = "xgrammar"
     disable_overlap = False
-    tp_size = 1
 
     @classmethod
     def setUpClass(cls):
@@ -36,9 +33,6 @@ class ServerWithGrammar(CustomTestCase):
             cls.backend,
         ]
 
-        if cls.tp_size > 1:
-            launch_args += ["--tp-size", str(cls.tp_size)]
-
         if cls.disable_overlap:
             launch_args += ["--disable-overlap-schedule"]
 
@@ -76,34 +70,5 @@ class TestLLGuidanceBackend(
     backend = "llguidance"
 
 
-# Test with TP=2
-@unittest.skipIf(torch.cuda.device_count() < 2, "Requires at least 2 GPUs")
-class TestXGrammarBackendTP2(
-    ServerWithGrammar,
-    TestJSONConstrainedMixin,
-    TestEBNFConstrainedMinxin,
-    TestRegexConstrainedMixin,
-):
-    backend = "xgrammar"
-    tp_size = 2
-
-
-@unittest.skipIf(torch.cuda.device_count() < 2, "Requires at least 2 GPUs")
-class TestOutlinesBackendTP2(ServerWithGrammar, TestJSONConstrainedMixin):
-    backend = "outlines"
-    tp_size = 2
-
-
-@unittest.skipIf(torch.cuda.device_count() < 2, "Requires at least 2 GPUs")
-class TestLLGuidanceBackendTP2(
-    ServerWithGrammar,
-    TestJSONConstrainedMixin,
-    TestEBNFConstrainedMinxin,
-    TestRegexConstrainedMixin,
-):
-    backend = "llguidance"
-    tp_size = 2
-
-
 if __name__ == "__main__":
     unittest.main()