Compare commits

...
Author SHA1 Message Date
sglang-bot bc35fd859e chore: update CI test est_time from recent run data 2026-11-16 00:00:30 +00:00
805 changed files with 825 additions and 825 deletions
@@ -15,7 +15,7 @@ from sglang.srt.layers.attention import aiter_mla_gluon as mod
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=15, suite="base-a-test-cpu")
register_cpu_ci(est_time=2, suite="base-a-test-cpu")
_GLUON_FN = "sglang.srt.layers.attention.aiter_mla_gluon._gluon_fn"
@@ -9,7 +9,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import CustomTestCase
# Triton kernel unit test for KV indices creation
register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=10, suite="stage-b-test-1-gpu-small-amd")
@@ -11,7 +11,7 @@ from sglang.test.test_utils import (
# Hybrid attention backend tests (FA3 prefill + FlashInfer decode, requires SM 90+ / H100)
# Multiple test classes: base, MLA, TorchCompile, SpecDecode variants
register_cuda_ci(est_time=393, stage="extra-a", runner_config="1-gpu-large")
register_cuda_ci(est_time=520, stage="extra-a", runner_config="1-gpu-large")
class TestHybridAttnBackendMLA(TestHybridAttnBackendBase):
@@ -12,7 +12,7 @@ from sglang.test.test_deterministic_utils import (
TestDeterministicBase,
)
register_cuda_ci(est_time=119, stage="extra-b", runner_config="4-gpu-h100")
register_cuda_ci(est_time=187, stage="extra-b", runner_config="4-gpu-h100")
QWEN35 = "Qwen/Qwen3.5-35B-A3B"
@@ -25,7 +25,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import CustomTestCase, is_in_amd_ci
# Triton attention kernel unit tests (decode, extend, prefill)
register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=36, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=30, suite="stage-b-test-1-gpu-small-amd")
@@ -17,7 +17,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
# trtllm_mha kernels are sm100-only; run this kernel-unit test on Blackwell.
register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=23, stage="base-b", runner_config="4-gpu-b200")
DEVICE = "cuda"
PAGE_SIZE = 32
@@ -25,7 +25,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import CustomTestCase
# Triton kernel unit test for the trtllm_mha device-side page-table build.
register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=13, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=14, stage="stage-b", runner_config="1-gpu-small-amd")
@@ -30,7 +30,7 @@ from sglang.kernels.ops.attention.verify_splitkv import (
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=15, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=30, suite="stage-b-test-1-gpu-small-amd-mi35x")
# Split-KV accumulates the prefix in parallel splits and merges via log-sum-exp;
@@ -36,7 +36,7 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=11, stage="base-a", runner_config="1-gpu-small")
register_cuda_ci(est_time=14, stage="base-a", runner_config="1-gpu-small")
_EXTEND_CASE = DenseAttentionCase(
name="extend_no_prefix_smoke",
@@ -30,7 +30,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=32, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -29,8 +29,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=27, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=22, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=37, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=35, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -26,8 +26,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=17, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=33, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=36, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(
@@ -14,7 +14,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=30, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=17, stage="base-b", runner_config="1-gpu-large")
@@ -18,8 +18,8 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=25, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=21, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(
@@ -21,8 +21,8 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=11, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=22, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -11,7 +11,7 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=26, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -30,7 +30,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=30, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=19, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=25, suite="stage-b-test-1-gpu-large-amd")
@@ -28,8 +28,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=17, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=32, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=88, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(
@@ -30,8 +30,8 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_draft_runner i
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=14, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=29, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=24, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -47,8 +47,8 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
run_dsv4_eagle_verify_cuda_graph_case,
)
register_cuda_ci(est_time=14, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=13, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=28, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=24, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -25,8 +25,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=15, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=24, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=31, stage="base-b", runner_config="1-gpu-large")
_cuda_major = int(torch.version.cuda.split(".")[0]) if torch.version.cuda else 0
_sm_major = torch.cuda.get_device_capability()[0] if torch.cuda.is_available() else 0
@@ -20,7 +20,7 @@ from sglang.kernels.ops.attention.fla.gdn_replayssm_spec_fold import (
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=19, stage="base-b", runner_config="1-gpu-large")
B, T = 3, 4
H, HV = 4, 8
@@ -15,7 +15,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=16, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -29,7 +29,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=13, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -22,7 +22,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=15, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=23, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=20, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -19,7 +19,7 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=17, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -32,7 +32,7 @@ from sglang.test.kits.attention_unittest.runner_modes.speculative_target_verify_
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=13, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=26, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
@@ -7,7 +7,7 @@ from sglang.srt.layers.attention.mamba.replay_state_indices_validator import (
)
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=6, suite="base-a-test-cpu")
register_cpu_ci(est_time=1, suite="base-a-test-cpu")
class TestReplayStateIndicesValidator(unittest.TestCase):
@@ -32,8 +32,8 @@ MLA_SHAPE_KWARGS = dict(
from sglang.test.ci.ci_register import register_cuda_ci
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=28, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=25, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -58,8 +58,8 @@ _DECODE_SKIP_REASON = (
from sglang.test.ci.ci_register import register_cuda_ci
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=27, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=26, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -28,8 +28,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=18, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=20, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=33, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=36, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(not torch.cuda.is_available(), "CUDA is required")
@@ -24,8 +24,8 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=13, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=28, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=24, stage="base-b", runner_config="1-gpu-large")
@unittest.skipIf(
@@ -24,7 +24,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardMode
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=11, suite="base-a-test-cpu")
register_cpu_ci(est_time=2, suite="base-a-test-cpu")
class TestSWAKVPoolSetKVBuffer(CustomTestCase):
@@ -12,7 +12,7 @@ from sglang.test.kits.attention_unittest.attention_methods.dense_attention impor
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=19, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -23,7 +23,7 @@ from sglang.test.kits.attention_unittest.runner_modes.split_op_runner import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=12, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=16, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
register_amd_ci(est_time=20, suite="stage-b-test-1-gpu-large-amd")
@@ -11,7 +11,7 @@ from sglang.srt.utils import get_cuda_driver_bindings, is_flashinfer_available
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=28, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=33, stage="base-b", runner_config="2-gpu-large")
WORLD_SIZE = 2
@@ -26,7 +26,7 @@ from sglang.benchmark.serving import (
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=16, suite="base-a-test-cpu")
register_cpu_ci(est_time=8, suite="base-a-test-cpu")
def _free_port() -> int:
@@ -60,7 +60,7 @@ from sglang.benchmark.serving import (
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=38, suite="base-a-test-cpu")
register_cpu_ci(est_time=14, suite="base-a-test-cpu")
register_cpu_ci(est_time=46, suite="stage-b-test-cpu-intel")
@@ -13,7 +13,7 @@ from sglang.benchmark.steady_state import (
)
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=10, suite="base-a-test-cpu")
register_cpu_ci(est_time=5, suite="base-a-test-cpu")
class _StringTokenizer:
@@ -25,7 +25,7 @@ from sglang.srt.multimodal.transport.cuda_ipc import (
from sglang.srt.runtime_context import get_context, get_parallel
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=12, suite="base-a-test-cpu")
register_cpu_ci(est_time=2, suite="base-a-test-cpu")
@pytest.fixture(autouse=True)
@@ -10,7 +10,7 @@ from sglang.test.scripted_runtime_chunked_helpers import (
run_until_finished,
)
register_cuda_ci(est_time=106, stage="extra-a", runner_config="1-gpu-small")
register_cuda_ci(est_time=112, stage="extra-a", runner_config="1-gpu-small")
register_amd_ci(est_time=198, stage="extra-a", runner_config="1-gpu-small-amd")
@@ -11,7 +11,7 @@ from sglang.test.scripted_runtime_chunked_helpers import (
run_until_finished,
)
register_cuda_ci(est_time=55, stage="extra-b", runner_config="4-gpu-h100")
register_cuda_ci(est_time=56, stage="extra-b", runner_config="4-gpu-h100")
_CHUNK_SIZE = 64
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=161, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=190, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=220, suite="stage-b-test-1-gpu-small-amd")
+1 -1
View File
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=97, stage="base-a", runner_config="1-gpu-small")
register_cuda_ci(est_time=121, stage="base-a", runner_config="1-gpu-small")
register_amd_ci(est_time=160, suite="stage-a-test-1-gpu-small-amd")
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=135, stage="base-a", runner_config="1-gpu-small")
register_cuda_ci(est_time=102, stage="base-a", runner_config="1-gpu-small")
class TestBasicSanityDFlash(
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=97, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=142, stage="base-b", runner_config="1-gpu-large")
TARGET_MODEL = "Qwen/Qwen3-14B"
DRAFT_MODEL = "deepseek-ai/dspark_qwen3_14b_block7"
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=167, stage="base-a", runner_config="1-gpu-small")
register_cuda_ci(est_time=131, stage="base-a", runner_config="1-gpu-small")
register_amd_ci(est_time=200, suite="stage-a-test-1-gpu-small-amd")
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
CustomTestCase,
)
register_cuda_ci(est_time=77, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=49, stage="base-b", runner_config="1-gpu-small")
class TestEngineChildPids(CustomTestCase):
+1 -1
View File
@@ -8,7 +8,7 @@ from sglang.srt.utils import get_device, is_hip
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST, CustomTestCase
register_cuda_ci(est_time=32, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=61, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=55, suite="stage-b-test-1-gpu-small-amd")
_is_hip = is_hip()
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=54, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=55, stage="base-b", runner_config="2-gpu-large")
class TestTPServerGPUProcesses(CustomTestCase):
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
send_generate_requests,
)
register_cuda_ci(est_time=53, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=76, stage="base-b", runner_config="1-gpu-small")
class TestMaxQueuedRequests(CustomTestCase):
+1 -1
View File
@@ -12,7 +12,7 @@ import pytest
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=7, suite="base-a-test-cpu")
register_cpu_ci(est_time=1, suite="base-a-test-cpu")
# Packages known to transitively depend on torch or triton.
# If a new package is added to runtime_base and it pulls torch,
+1 -1
View File
@@ -31,7 +31,7 @@ from sglang.test.test_utils import (
run_logprob_check,
)
register_cuda_ci(est_time=265, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=270, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=260, suite="stage-b-test-1-gpu-small-amd")
SERVER_ENV = {"SGLANG_USE_PICKLE_IPC": "0"}
+1 -1
View File
@@ -41,7 +41,7 @@ from sglang.srt.runtime_context import get_parallel
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=11, suite="base-a-test-cpu")
register_cpu_ci(est_time=3, suite="base-a-test-cpu")
class _ExtendMode:
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
try_cached_model,
)
register_cuda_ci(est_time=689, stage="extra-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=802, stage="extra-b", runner_config="4-gpu-b200")
MODEL = "deepseek-ai/DeepSeek-V4-Flash"
SERVER_LAUNCH_TIMEOUT = 3600
@@ -12,7 +12,7 @@ import unittest
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.server_fixtures.pcg_spec_fixture import PCGSpecBase
register_cuda_ci(est_time=106, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=110, stage="base-b", runner_config="2-gpu-large")
class TestBCGWithEAGLE3(PCGSpecBase, unittest.TestCase):
+1 -1
View File
@@ -33,7 +33,7 @@ from sglang.srt.mem_cache.memory_pool import HybridLinearKVPool
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=11, suite="base-a-test-cpu")
register_cpu_ci(est_time=4, suite="base-a-test-cpu")
DCP_SIZES = [1, 2, 3, 4, 8]
LENS = list(range(0, 41))
@@ -23,7 +23,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardMode
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=10, stage="base-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=16, stage="base-b", runner_config="4-gpu-b200")
NUM_DRAFT_TOKENS = 8
DCP_SIZE = 4
@@ -18,7 +18,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
from sglang.test.layer_ut_utils import init_single_process_dist
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-small")
# Backend-specific: the hook resolves get_device() and the matching distributed
# backend, so only an AMD run exercises the HIP device and dump path.
register_amd_ci(est_time=15, suite="stage-b-test-1-gpu-small-amd")
@@ -15,7 +15,7 @@ from sglang.test.test_utils import (
QWEN3_8B_MODEL_PATH = "Qwen/Qwen3-8B"
register_cuda_ci(est_time=125, stage="base-c", runner_config="4-gpu-gb300")
register_cuda_ci(est_time=223, stage="base-c", runner_config="4-gpu-gb300")
# Keep rendezvous ports below the ephemeral range on the 4-GPU GB300 runner.
NCCL_PORT_BASE = DEFAULT_PORT_FOR_SRT_TEST_RUNNER + 120
@@ -27,7 +27,7 @@ from sglang.test.test_utils import (
DEFAULT_TARGET_MODEL_EAGLE3,
)
register_cuda_ci(est_time=1007, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=671, stage="base-b", runner_config="2-gpu-large")
class TestDisaggregationAccuracy(PauseResumeInPlaceMixin, PDDisaggregationServerBase):
@@ -12,7 +12,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
)
from sglang.test.test_utils import DEFAULT_SMALL_MODEL_NAME_FOR_TEST
register_cuda_ci(est_time=120, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=82, stage="base-b", runner_config="2-gpu-large")
PD_CHUNKED_ABORT_EXTRA_ARGS = [
"--max-running-requests",
@@ -17,7 +17,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
)
from sglang.test.test_utils import DEFAULT_MODEL_NAME_FOR_TEST_MXFP4_WITH_MOE, is_in_ci
register_cuda_ci(est_time=365, stage="extra-b", runner_config="8-gpu-h200")
register_cuda_ci(est_time=171, stage="extra-b", runner_config="8-gpu-h200")
SWA_SERVER_ARGS = ["--page-size", "64", "--attention-backend", "triton"]
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
try_cached_model,
)
register_cuda_ci(est_time=212, stage="extra-b", runner_config="8-gpu-h200")
register_cuda_ci(est_time=213, stage="extra-b", runner_config="8-gpu-h200")
DSV4_FLASH_MODEL = "sgl-project/DeepSeek-V4-Flash-FP8"
DSV4_FLASH_LOADER_CONFIG = '{"enable_multithread_load": true, "num_threads": 64}'
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
popen_launch_pd_server,
)
register_cuda_ci(est_time=578, stage="extra-b", runner_config="8-gpu-h200")
register_cuda_ci(est_time=575, stage="extra-b", runner_config="8-gpu-h200")
@unittest.skipIf(is_in_ci(), "Temporarily disable the flaky test.")
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
try_cached_model,
)
register_cuda_ci(est_time=447, stage="extra-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=624, stage="extra-b", runner_config="4-gpu-b200")
MODEL = os.environ.get(
"INKLING_SMALL_TEST_MODEL_PATH", "thinkingmachines/Inkling-Small-NVFP4"
@@ -27,7 +27,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
)
from sglang.test.test_utils import DEFAULT_MODEL_NAME_FOR_TEST, is_rust_server_built
register_cuda_ci(est_time=66, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=92, stage="base-b", runner_config="2-gpu-large")
@unittest.skipUnless(
+1 -1
View File
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=443, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=330, stage="base-b", runner_config="2-gpu-large")
register_amd_ci(est_time=500, suite="stage-b-test-2-gpu-large-amd")
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=84, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=97, stage="base-b", runner_config="2-gpu-large")
register_amd_ci(est_time=73, suite="stage-b-test-2-gpu-large-amd")
@@ -29,7 +29,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
)
from sglang.test.test_utils import DEFAULT_MODEL_NAME_FOR_TEST_MXFP4_WITH_MOE
register_cuda_ci(est_time=1200, stage="extra-a", runner_config="2-gpu-large")
register_cuda_ci(est_time=505, stage="extra-a", runner_config="2-gpu-large")
UNIFIED_SWA_ARGS = [
"--skip-tokenizer-init",
@@ -32,7 +32,7 @@ from sglang.test.server_fixtures.disaggregation_fixture import (
PDDisaggregationServerBase,
)
register_cuda_ci(est_time=900, stage="extra-a", runner_config="2-gpu-large")
register_cuda_ci(est_time=168, stage="extra-a", runner_config="2-gpu-large")
_MODEL_PATH = os.environ.get("INKLING_TEST_MODEL_PATH", "thinkingmachines/Inkling")
_MODEL_REVISION = os.environ.get("INKLING_TEST_MODEL_REVISION", "test")
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=600, stage="extra-a", runner_config="2-gpu-large")
register_cuda_ci(est_time=478, stage="extra-a", runner_config="2-gpu-large")
_COMMON_ARGS = [
"--trust-remote-code",
@@ -15,7 +15,7 @@ from sglang.srt.layers.logprob_processor import InputLogprobProcessor
from sglang.srt.model_executor.runner_utils import pool
from sglang.test.ci.ci_register import register_cuda_ci
register_cuda_ci(est_time=60, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=8, stage="base-b", runner_config="2-gpu-large")
def _run_rank(rank, rendezvous):
@@ -27,7 +27,7 @@ from sglang.test.lora_utils import (
)
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=600, stage="extra-a", runner_config="1-gpu-large")
register_cuda_ci(est_time=159, stage="extra-a", runner_config="1-gpu-large")
# Missing adapters shift logprobs by 7-17.
LOGPROB_THRESHOLD = 1.0
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
try_cached_model,
)
register_cuda_ci(est_time=283, stage="extra-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=340, stage="extra-b", runner_config="4-gpu-b200")
MODEL = "deepseek-ai/DeepSeek-V4-Flash-0731"
SERVER_LAUNCH_TIMEOUT = 3600
@@ -1,7 +1,7 @@
from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
# Generation model tests (CUDA only)
register_cuda_ci(est_time=144, stage="extra-a", runner_config="1-gpu-large")
register_cuda_ci(est_time=229, stage="extra-a", runner_config="1-gpu-large")
register_amd_ci(est_time=106, suite="stage-b-test-1-gpu-small-amd")
# Copyright 2023-2024 SGLang Team
@@ -48,7 +48,7 @@ from sglang.test.test_utils import (
terminate_and_kill_process_tree,
)
register_cuda_ci(est_time=1079, stage="extra-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=911, stage="extra-b", runner_config="4-gpu-b200")
_MODEL_PATH = os.environ.get(
"INKLING_SMALL_TEST_MODEL_PATH", "thinkingmachines/Inkling-Small-NVFP4"
@@ -17,7 +17,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=1472, stage="extra-b", runner_config="8-gpu-b300")
register_cuda_ci(est_time=2287, stage="extra-b", runner_config="8-gpu-b300")
MODEL_PATH = "moonshotai/Kimi-K3"
DSPARK_DRAFT_MODEL = "RadixArk/Kimi-K3-DSpark"
@@ -47,7 +47,7 @@ from sglang.test.kits.eval_accuracy_kit import GSM8KMixin
from sglang.test.kits.prefix_cache_branching_kit import PrefixCacheBranchingMixin
from sglang.test.server_fixtures.default_fixture import DefaultServerBase
register_cuda_ci(est_time=161, stage="extra-b", runner_config="4-gpu-b200")
register_cuda_ci(est_time=233, stage="extra-b", runner_config="4-gpu-b200")
KIMI_LINEAR_MODEL = "moonshotai/Kimi-Linear-48B-A3B-Instruct"
@@ -20,7 +20,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=74, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=77, stage="base-b", runner_config="2-gpu-large")
# Qwen3 dense (architecture "Qwen3ForCausalLM") is the SP allowlist entry.
LAYERNORM_SP_MODEL = "Qwen/Qwen3-8B"
@@ -11,7 +11,7 @@ from sglang.test.ci.ci_register import (
from sglang.test.few_shot_gsm8k import run_eval
from sglang.test.server_fixtures.default_fixture import DefaultServerBase
register_cuda_ci(est_time=66, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=80, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=48, suite="stage-b-test-1-gpu-small-amd")
register_cpu_ci(est_time=206, suite="stage-b-test-cpu-intel")
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=217, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=234, stage="base-b", runner_config="1-gpu-small")
register_amd_ci(est_time=320, suite="stage-b-test-1-gpu-small-amd")
+1 -1
View File
@@ -21,7 +21,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=900, stage="extra-b", runner_config="4-gpu-h100")
register_cuda_ci(est_time=423, stage="extra-b", runner_config="4-gpu-h100")
# topk=1 chains and a topk=2 tree: the relayed topology is constant for the
# former and data-dependent for the latter, so both shapes are covered.
@@ -16,7 +16,7 @@ from sglang.test.test_utils import (
publish_build_topology,
)
register_cuda_ci(est_time=60, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=16, stage="base-b", runner_config="1-gpu-small")
# Shrunk from inclusionAI/Ling-mini-2.0; the field names are the real
# checkpoint's, only the sizes are cut down for test speed.
@@ -14,7 +14,7 @@ from sglang.test.test_utils import (
popen_launch_server,
)
register_cuda_ci(est_time=600, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=149, stage="base-b", runner_config="2-gpu-large")
class TestDFlashDominoFullVocab(CustomTestCase):
@@ -13,7 +13,7 @@ from sglang.test.test_utils import (
terminate_and_kill_process_tree,
)
register_cuda_ci(est_time=283, stage="extra-b", runner_config="4-gpu-h100")
register_cuda_ci(est_time=287, stage="extra-b", runner_config="4-gpu-h100")
class TestHybridDPTP(CustomTestCase):
@@ -12,7 +12,7 @@ from sglang.test.test_utils import (
terminate_and_kill_process_tree,
)
register_cuda_ci(est_time=253, stage="extra-b", runner_config="8-gpu-h200")
register_cuda_ci(est_time=387, stage="extra-b", runner_config="8-gpu-h200")
DEEPSEEK_V3_MODEL_PATH = "deepseek-ai/DeepSeek-V3-0324"
@@ -40,7 +40,7 @@ from sglang.srt.distributed.parallel_state import (
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import publish_build_topology
register_cuda_ci(est_time=18, stage="base-b", runner_config="2-gpu-large")
register_cuda_ci(est_time=24, stage="base-b", runner_config="2-gpu-large")
NUM_GPUS = 2
TOPK = 2
@@ -13,7 +13,7 @@ import torch
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=10, suite="base-a-test-cpu")
register_cpu_ci(est_time=1, suite="base-a-test-cpu")
from sglang.srt.layers.moe.expert_pack import ( # noqa: E402
ExpertPackStore,
@@ -20,7 +20,7 @@ from sglang.srt.model_loader.kimi_k3_gguf import (
from sglang.srt.models.kimi_k3 import _uses_split_gguf_kv_b
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=11, suite="base-a-test-cpu")
register_cpu_ci(est_time=2, suite="base-a-test-cpu")
class TestKimiK3GGUFMapping(unittest.TestCase):
@@ -16,7 +16,7 @@ from sglang.srt.function_call.kimik3_format import (
)
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=7, suite="base-a-test-cpu")
register_cpu_ci(est_time=2, suite="base-a-test-cpu")
def _make_tool(name: str) -> Tool:
@@ -17,7 +17,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.hicache_spec_storage_common import HiCacheSpecStorageMixin
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=142, stage="extra-a", runner_config="1-gpu-large")
register_cuda_ci(est_time=224, stage="extra-a", runner_config="1-gpu-large")
@unittest.skipIf(is_hip(), "HiCache + EAGLE3 file-storage loadback e2e is CUDA-only.")
@@ -29,7 +29,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.hicache_spec_storage_common import HiCacheSpecStorageMixin
from sglang.test.test_utils import CustomTestCase, find_available_port
register_cuda_ci(est_time=186, stage="extra-a", runner_config="2-gpu-large")
register_cuda_ci(est_time=190, stage="extra-a", runner_config="2-gpu-large")
@unittest.skipIf(
@@ -22,7 +22,7 @@ from sglang.test.test_utils import (
terminate_and_kill_process_tree,
)
register_cuda_ci(est_time=369, stage="extra-b", runner_config="4-gpu-h100")
register_cuda_ci(est_time=295, stage="extra-b", runner_config="4-gpu-h100")
QWEN35_27B_MODEL = "Qwen/Qwen3.5-27B"
ACC_THRESHOLDS = {QWEN35_27B_MODEL: {"gsm8k": 0.8}}
@@ -18,7 +18,7 @@ from sglang.srt.layers.quantization.fp8_utils import (
)
from sglang.test.ci.ci_register import register_cuda_ci
register_cuda_ci(est_time=90, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-small")
def _make_inputs(m: int, n: int, k: int):
@@ -16,7 +16,7 @@ import torch
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=23, stage="base-b", runner_config="1-gpu-large")
class TestLSECombineTritonVsCPU(CustomTestCase):
@@ -34,7 +34,7 @@ from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMo
from sglang.srt.server_args import ServerArgs, set_global_server_args_for_scheduler
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")
register_cuda_ci(est_time=31, stage="base-b", runner_config="1-gpu-large")
# Global configuration for all indexer tests
DEFAULT_CONFIG = {
@@ -29,7 +29,7 @@ from sglang.srt.layers.attention.dsv4.indexer import (
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-small")
register_cuda_ci(est_time=20, stage="base-b", runner_config="1-gpu-small")
# DSv4 indexer cache layout (fixed by deepseek_v4_memory_pool.DeepSeekV4IndexerPool):
@@ -35,7 +35,7 @@ from sglang.srt.runtime_context import get_parallel
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.kernels.utils import multigpu_pytest_main
register_cuda_ci(est_time=38, stage="extra-b", runner_config="8-gpu-h200")
register_cuda_ci(est_time=32, stage="extra-b", runner_config="8-gpu-h200")
# Nightly is not redundant here: it sets SGLANG_JIT_KERNEL_RUN_FULL_TESTS=1 to expand get_ci_test_range sweeps.
register_cuda_ci(est_time=70, stage="nightly", runner_config="8-gpu-h200")
@@ -49,7 +49,7 @@ from sglang.kernels.ops.diffusion import (
from sglang.test.ci.ci_register import register_cpu_ci, register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=12, suite="base-a-test-cpu")
register_cpu_ci(est_time=8, suite="base-a-test-cpu")
register_cuda_ci(est_time=38, stage="base-b-kernel-unit", runner_config="1-gpu-large")
requires_cuda = pytest.mark.skipif(

Some files were not shown because too many files have changed in this diff Show More