[Spec] Async-assert probes across EAGLE/MTP; zero tgt_cache_loc (#26335)

This commit is contained in:
Liangsheng Yin
2026-05-26 11:34:03 -07:00
committed by GitHub
parent 6c8128650e
commit ec6f8d61f7
23 changed files with 171 additions and 94 deletions
+1 -2
View File
@@ -55,8 +55,7 @@ class TestDFlashServerBase(CustomTestCase, MatchedStopMixin, GSM8KMixin):
try:
with (
envs.SGLANG_ENABLE_STRICT_MEM_CHECK_DURING_BUSY.override(1),
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True),
):
cls.process = popen_launch_server(
cls.model,
@@ -49,10 +49,7 @@ class TestDeepseekV3FP4MTP(CustomTestCase):
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
]
with (
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
):
with envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True):
cls.process = popen_launch_server(
cls.model,
cls.base_url,
@@ -62,8 +62,7 @@ class TestEagleConstrainedDecoding(
launch_args.extend(cls.other_launch_args)
with (
envs.SGLANG_ENABLE_SPEC_V2.override(cls.spec_v2),
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True),
):
cls.process = popen_launch_server(
cls.model,
@@ -60,10 +60,7 @@ class TestEAGLE3EngineDPAttention(CustomTestCase):
"--cuda-graph-max-bs",
"64",
]
with (
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
):
with envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True):
cls.process = popen_launch_server(
cls.model,
cls.base_url,
@@ -65,8 +65,7 @@ class TestEagle3ServerBase(CustomTestCase, MatchedStopMixin):
launch_args.extend(cls.other_launch_args)
with (
envs.SGLANG_ENABLE_STRICT_MEM_CHECK_DURING_BUSY.override(1),
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True),
envs.SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN.override(True),
):
cls.process = popen_launch_server(
@@ -65,10 +65,7 @@ class TestEagleDPAttnServerSmall(CustomTestCase):
"--speculative-num-draft-tokens",
"4",
]
with (
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
):
with envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True):
cls.process = popen_launch_server(
cls.model,
cls.base_url,
@@ -73,10 +73,7 @@ class TestEagleDPAttnServerLarge(CustomTestCase):
"--model-loader-extra-config",
'{"enable_multithread_load": true,"num_threads": 64}',
]
with (
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
):
with envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True):
cls.process = popen_launch_server(
cls.model,
cls.base_url,
@@ -51,10 +51,7 @@ class ServerWithGrammar(CustomTestCase):
"--speculative-num-draft-tokens=8",
]
with (
envs.SGLANG_SPEC_NAN_DETECTION.override(True),
envs.SGLANG_SPEC_OOB_DETECTION.override(True),
):
with envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True):
cls.process = popen_launch_server(
cls.model,
cls.base_url,