chore: update CI test est_time values (#38238)

Co-authored-by: sglang-bot <sglang-bot@users.noreply.github.com>
2026-09-06 17:49:41 -07:00
parent 15aa2fb843
commit 6252993afe
1016 changed files with 1051 additions and 1051 deletions
@@ -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=30, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=14, stage="base-b", runner_config="1-gpu-large")

 # Global configuration for all indexer tests
 DEFAULT_CONFIG = {
@@ -10,7 +10,7 @@ from sglang.kernels.ops.attention.dsa_metadata 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=15, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=15, stage="stage-b", runner_config="1-gpu-large-amd")


@@ -11,7 +11,7 @@ from sglang.kernels.ops.attention.dsa.transform_index import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase

-register_cuda_ci(est_time=60, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-large")

 TOPK = 2048

@@ -31,7 +31,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

 _is_hip = is_hip()

-register_cuda_ci(est_time=45, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=45, suite="jit-kernel-unit-test-amd")

 HEAD_DIM = 128
@@ -28,7 +28,7 @@ from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

 _is_hip = is_hip()

-register_cuda_ci(est_time=45, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=13, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=45, suite="jit-kernel-unit-test-amd")

 HEAD_DIM = 128
@@ -53,7 +53,7 @@ from sglang.srt.runtime_context import get_resources
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase

-register_cuda_ci(est_time=45, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-large")


 # Per-token byte layout
@@ -14,7 +14,7 @@ from sglang.kernels.ops.attention.triton_gdn_fused_proj import (
 from sglang.kernels.ops.mamba.causal_conv1d_triton import causal_conv1d_update
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=7, stage="base-b", runner_config="1-gpu-large")


 def _reference(
@@ -14,7 +14,7 @@ from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import CustomTestCase

 register_cuda_ci(est_time=180, stage="base-b-kernel-unit", runner_config="4-gpu-b200")
-register_cuda_ci(est_time=180, stage="base-c", runner_config="4-gpu-gb300")
+register_cuda_ci(est_time=80, stage="base-c", runner_config="4-gpu-gb300")


 def _inputs(seed, seq_len=128):
@@ -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=20, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=11, stage="base-b", runner_config="1-gpu-small")


 # DSv4 indexer cache layout (fixed by deepseek_v4_memory_pool.DeepSeekV4IndexerPool):
@@ -34,7 +34,7 @@ from sglang.srt.distributed.device_communicators.triton_symm_mem_ag import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kernels.utils import multigpu_pytest_main

-register_cuda_ci(est_time=240, stage="extra-b", runner_config="8-gpu-h200")
+register_cuda_ci(est_time=38, 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")

@@ -26,7 +26,7 @@ from sglang.kernels.ops.diffusion import _EXPORTS, _SPECS
 from sglang.kernels.registry import registry
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=25, suite="base-a-test-cpu")
+register_cpu_ci(est_time=16, suite="base-a-test-cpu")

 PACKAGE = "sglang.kernels.ops.diffusion"
 _PACKAGE_DIR = pathlib.Path(importlib.import_module(PACKAGE).__file__ or "").parent
@@ -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=3, suite="base-a-test-cpu")
+register_cpu_ci(est_time=12, 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(
@@ -14,7 +14,7 @@ from sglang.srt.layers.vocab_parallel_embedding import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=13, stage="base-b", runner_config="1-gpu-small")

 pytestmark = pytest.mark.skipif(
    not torch.cuda.is_available(), reason="CUDA is required for this test."
@@ -23,7 +23,7 @@ from sglang.srt.distributed.device_communicators.custom_all_reduce_v2 import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kernels.utils import multigpu_pytest_main

-register_cuda_ci(est_time=240, stage="base-c", runner_config="4-gpu-b200")
+register_cuda_ci(est_time=34, stage="base-c", runner_config="4-gpu-b200")
 register_cuda_ci(est_time=480, stage="nightly", runner_config="8-gpu-b200")

 _HIDDEN_SIZE = 7168
@@ -8,7 +8,7 @@ from sglang.kernels.jit.benchmark.kv_canary.utils import (
 )
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=3, stage="base-a", runner_config="cpu")
+register_cpu_ci(est_time=7, stage="base-a", runner_config="cpu")


 def test_fast_matrix_cases_include_e2e_decode_and_chunked_prefill_scenarios() -> None:
@@ -14,7 +14,7 @@ from sglang.srt.mem_cache.pool_host.mla import MLATokenToKVPoolHost
 from sglang.srt.utils import is_cuda, is_hip, is_npu, is_xpu
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")

 pytestmark = pytest.mark.skipif(
    not torch.cuda.is_available()
@@ -24,7 +24,7 @@ from sglang.srt.mem_cache.pool_host.mla import MLATokenToKVPoolHost
 from sglang.srt.utils import is_cuda, is_hip, is_npu, is_xpu
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=15, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=30, stage="jit-kernel-unit", runner_config="amd")

 pytestmark = pytest.mark.skipif(
@@ -7,7 +7,7 @@ import torch
 from sglang.srt.mem_cache.pool_host.mha import AsymmetricMHATokenToKVPoolHost
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=9, stage="base-b", runner_config="1-gpu-large")
 register_amd_ci(est_time=10, suite="nightly-amd-kernel-1-gpu", nightly=True)

 # These tests use AsymmetricMHATokenToKVPoolHost methods and let that class call
@@ -16,7 +16,7 @@ from sglang.kernels.spec import CapabilityRequirement as Cap
 from sglang.kernels.spec import KernelBackend, KernelSpec
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=30, suite="base-a-test-cpu")
+register_cpu_ci(est_time=7, suite="base-a-test-cpu")


 class _ToyAdd(BaseFusedOp):
@@ -11,7 +11,7 @@ import torch
 from sglang.kernels.spec import KernelBackend
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=60, stage="extra-a", runner_config="1-gpu-small")
+register_cuda_ci(est_time=12, stage="extra-a", runner_config="1-gpu-small")

 # torch_compile is native under the hood; skip it here (compile time dominates)
 # -- it is exercised in the CPU lane.
@@ -16,7 +16,7 @@ import torch

 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=60, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-small")

 pytestmark = pytest.mark.skipif(
    not torch.cuda.is_available(),
@@ -15,7 +15,7 @@ from sglang.kernels import DeviceType, KernelBackend, PlatformInfo
 from sglang.kernels.spec import CapabilityRequirement as Cap
 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=24, suite="base-a-test-cpu")

 GROUPS = K.ops.__all__

@@ -7,7 +7,7 @@ import sglang.kernels.ops.layernorm.mhc as mhc
 from sglang.kernels.ops.layernorm.mhc import mhc_fused_post_pre, mhc_post, mhc_pre
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=30, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")


@pytest.mark.parametrize("hidden_size", [4096, 7168])
@@ -20,7 +20,7 @@ from sglang.test.test_marlin_utils import (
    marlin_quantize,
 )

-register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-large")


 def _has_aot_moe_wna16_marlin_gemm() -> bool:
@@ -19,8 +19,8 @@ from sglang.srt.utils.common import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import make_nvfp4_weight_and_ref

-register_cuda_ci(est_time=6, stage="base-b", runner_config="1-gpu-large")
-register_cuda_ci(est_time=6, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=12, stage="base-b", runner_config="1-gpu-small")


@pytest.mark.skipif(
@@ -1,6 +1,6 @@
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=15, stage="base-b", runner_config="1-gpu-small")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-small")

 """Boundary-KV fix kernels (SGLANG_ENABLE_MTP_BOUNDARY_KV_FIX) vs a pure-torch reference.

@@ -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=30, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=10, stage="base-b", runner_config="1-gpu-large")


 class TestLSECombineTritonVsCPU(CustomTestCase):
@@ -14,7 +14,7 @@ from sglang.kernels.ops.mamba.causal_conv1d_triton import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=60, stage="base-b", runner_config="1-gpu-large")
+register_cuda_ci(est_time=8, stage="base-b", runner_config="1-gpu-large")

 _DEVICE = "cuda"

@@ -28,7 +28,7 @@ from sglang.kernels.spec import CapabilityRequirement as Cap
 from sglang.kernels.spec import KernelBackend, PlatformInfo
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=60, suite="base-a-test-cpu")
+register_cpu_ci(est_time=12, suite="base-a-test-cpu")

 _CUDA = PlatformInfo(device_type="cuda", cuda_arch_major=9, cuda_arch_minor=0)
 _HIP = PlatformInfo(device_type="hip")
@@ -19,7 +19,7 @@ from sglang.kernels.jit.utils.compile.paths import KERNEL_PATH
 from sglang.kernels.jit.utils.compile.spec import BuildSpec
 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=8, suite="base-a-test-cpu")


@pytest.fixture(autouse=True)
@@ -12,7 +12,7 @@ import pytest
 import sglang.kernels as kernels
 from sglang.test.ci.ci_register import register_cpu_ci

-register_cpu_ci(est_time=5, suite="base-a-test-cpu")
+register_cpu_ci(est_time=13, suite="base-a-test-cpu")

 REPO_ROOT = Path(__file__).resolve().parents[3]
 KERNELS_ROOT = REPO_ROOT / "python" / "sglang" / "kernels"