[Spec] Retire Spec V1 (#27964)

test/registered/spec/dflash/test_dflash.py

@@ -28,7 +28,8 @@ class TestDFlashServerBase(CustomTestCase, MatchedStopMixin, GSM8KMixin):
     attention_backend = "flashinfer"
     page_size = 1
     other_launch_args = []
-    spec_v2 = False
+    # Base classes exercise the non-overlap (synchronous) scheduling path.
+    disable_overlap = True
     overlap_plan_stream = False
     model = DEFAULT_TARGET_MODEL_DFLASH
     draft_model = DEFAULT_DRAFT_MODEL_DFLASH
@@ -57,9 +58,10 @@ class TestDFlashServerBase(CustomTestCase, MatchedStopMixin, GSM8KMixin):
             "--cuda-graph-bs",
             *[str(i) for i in range(1, cls.max_running_requests + 1)],
         ]
+        if cls.disable_overlap:
+            launch_args.append("--disable-overlap-schedule")
         launch_args.extend(cls.other_launch_args)
         with (
-            envs.SGLANG_ENABLE_SPEC_V2.override(cls.spec_v2),
             envs.SGLANG_ENABLE_OVERLAP_PLAN_STREAM.override(cls.overlap_plan_stream),
             envs.SGLANG_ENABLE_STRICT_MEM_CHECK_DURING_BUSY.override(1),
             envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True),
@@ -150,7 +152,7 @@ class TestDFlashServerNoCudaGraph(TestDFlashServerBase):
 
 
 class TestDFlashServerSpecV2(TestDFlashServerBase):
-    spec_v2 = True
+    disable_overlap = False
 
     def test_radix_attention(self):
         run_radix_attention_test(self.base_url)

test/registered/spec/eagle/test_eagle_constrained_decoding.py

@@ -1,6 +1,5 @@
 import unittest
 
-from sglang.srt.environ import envs
 from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kits.json_constrained_kit import JSONConstrainedMixin
@@ -31,7 +30,8 @@ class TestEagleConstrainedDecoding(
     model = DEFAULT_TARGET_MODEL_EAGLE
     draft_model = DEFAULT_DRAFT_MODEL_EAGLE
     grammar_backend = "xgrammar"
-    spec_v2 = False
+    # Run the synchronous (non-overlap) scheduling path.
+    disable_overlap = True
 
     @classmethod
     def setUpClass(cls):
@@ -59,14 +59,15 @@ class TestEagleConstrainedDecoding(
             "--grammar-backend",
             cls.grammar_backend,
         ]
+        if cls.disable_overlap:
+            launch_args.append("--disable-overlap-schedule")
         launch_args.extend(cls.other_launch_args)
-        with envs.SGLANG_ENABLE_SPEC_V2.override(cls.spec_v2):
-            cls.process = popen_launch_server(
-                cls.model,
-                cls.base_url,
-                timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-                other_args=launch_args,
-            )
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=launch_args,
+        )
 
     @classmethod
     def tearDownClass(cls):
@@ -74,7 +75,7 @@ class TestEagleConstrainedDecoding(
 
 
 class TestEagleConstrainedDecodingV2(TestEagleConstrainedDecoding):
-    spec_v2 = True
+    disable_overlap = False
 
 
 if __name__ == "__main__":

test/registered/spec/test_spec_standalone_extra.py

@@ -13,14 +13,14 @@ class TestStandaloneSpeculativeDecodingBase(StandaloneServerBase, CustomTestCase
     attention_backend = "fa3"
     speculative_eagle_topk = 2
     speculative_num_draft_tokens = 7
-    enable_spec_v2 = False
+    disable_overlap = True
 
 
 class TestStandaloneSpeculativeDecodingTriton(StandaloneServerBase, CustomTestCase):
     attention_backend = "triton"
     speculative_eagle_topk = 2
     speculative_num_draft_tokens = 7
-    enable_spec_v2 = False
+    disable_overlap = True
     enable_deterministic_inference = True
 
 
@@ -28,7 +28,7 @@ class TestStandaloneSpeculativeDecodingFlashinfer(StandaloneServerBase, CustomTe
     attention_backend = "flashinfer"
     speculative_eagle_topk = 2
     speculative_num_draft_tokens = 7
-    enable_spec_v2 = False
+    disable_overlap = True
 
 
 if __name__ == "__main__":