[CI] Rename basic CI stage-a/b/c -> base-a/b/c for symmetry with extra CI (#25420)

2026-05-15 18:26:55 -07:00
parent d0c38329b2
commit b7d62bd724
473 changed files with 745 additions and 746 deletions
@@ -19,7 +19,7 @@ from sglang.jit_kernel.benchmark.utils import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=30, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-benchmark-1-gpu-large")


@torch.compile
@@ -9,7 +9,7 @@ from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.utils import is_in_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")

 try:
    from sgl_kernel import awq_dequantize as aot_awq_dequantize
@@ -10,7 +10,7 @@ from sglang.jit_kernel.benchmark.utils import (
 from sglang.jit_kernel.cast import downcast_fp8 as downcast_fp8_jit
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=10, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-benchmark-1-gpu-large")

 DEVICE = DEFAULT_DEVICE
 DTYPE = torch.bfloat16
@@ -13,7 +13,7 @@ from sglang.jit_kernel.clamp_position import clamp_position_cuda
 from sglang.srt.utils import get_compiler_backend
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=13, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=13, suite="base-b-kernel-benchmark-1-gpu-large")
 register_amd_ci(est_time=16, suite="jit-kernel-unit-test-amd")

 SIZE_LIST = get_benchmark_range(
@@ -12,7 +12,7 @@ from sglang.jit_kernel.concat_mla import concat_mla_k as jit_k
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.utils import is_in_ci

-register_cuda_ci(est_time=6, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=6, suite="base-b-kernel-benchmark-1-gpu-large")

 IS_CI = is_in_ci()

@@ -26,7 +26,7 @@ from sglang.test.ci.ci_register import register_cuda_ci

 register_cuda_ci(
    est_time=120,
-    suite="stage-b-kernel-benchmark-1-gpu-large",
+    suite="base-b-kernel-benchmark-1-gpu-large",
    disabled="requires multi-GPU, self-skips in CI",
 )

@@ -20,7 +20,7 @@ from sglang.jit_kernel.fused_qknorm_rope import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=6, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=6, suite="base-b-kernel-benchmark-1-gpu-large")

 try:
    from sgl_kernel import fused_qk_norm_rope as fused_qk_norm_rope_aot
@@ -16,7 +16,7 @@ from sglang.jit_kernel.benchmark.utils import (
 from sglang.jit_kernel.hadamard import hadamard_transform
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")

 # AOT kernel: might not be available in all environments.
 # This is used for performance baseline comparison.
@@ -33,7 +33,7 @@ from sglang.jit_kernel.hicache import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=29, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=29, suite="base-b-kernel-benchmark-1-gpu-large")

 DISABLE_TORCH = os.environ.get("DISABLE_TORCH", "0") == "1"
 PAGE_SIZE = 1
@@ -9,7 +9,7 @@ from sglang.jit_kernel.benchmark.utils import DEFAULT_DEVICE, DEFAULT_DTYPE
 from sglang.jit_kernel.hisparse import load_cache_to_device_buffer_mla
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=12, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=12, suite="base-b-kernel-benchmark-1-gpu-large")

 DEVICE = DEFAULT_DEVICE
 DTYPE = DEFAULT_DTYPE
@@ -20,7 +20,7 @@ from sglang.jit_kernel.mla_kv_pack_quantize_fp8 import (
 from sglang.jit_kernel.utils import is_arch_support_pdl
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=15, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=15, suite="base-b-kernel-benchmark-1-gpu-large")


@triton.jit
@@ -13,7 +13,7 @@ from sglang.jit_kernel.mxfp8 import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")


 def is_sm100_supported(device=None) -> bool:
@@ -11,7 +11,7 @@ from sglang.jit_kernel.norm import fused_add_rmsnorm as jit_fused_add_rmsnorm
 from sglang.jit_kernel.norm import rmsnorm as jit_rmsnorm
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=30, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-benchmark-1-gpu-large")


 DTYPE = torch.bfloat16
@@ -15,7 +15,7 @@ from sglang.jit_kernel.nvfp4 import (
 from sglang.srt.utils import is_sm100_supported
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")

 FLOAT4_E2M1_MAX = 6.0
 FLOAT8_E4M3_MAX = torch.finfo(torch.float8_e4m3fn).max
@@ -10,7 +10,7 @@ from sglang.jit_kernel.nvfp4 import scaled_fp4_quant
 from sglang.srt.utils import is_sm100_supported
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")

 FLOAT4_E2M1_MAX = 6.0
 FLOAT8_E4M3_MAX = torch.finfo(torch.float8_e4m3fn).max
@@ -10,7 +10,7 @@ from sglang.jit_kernel.nvfp4 import cutlass_scaled_fp4_mm, scaled_fp4_quant
 from sglang.srt.utils import is_sm100_supported, is_sm120_supported
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")

 FLOAT4_E2M1_MAX = 6.0
 FLOAT8_E4M3_MAX = torch.finfo(torch.float8_e4m3fn).max
@@ -8,7 +8,7 @@ from sglang.jit_kernel.benchmark.utils import get_benchmark_range, run_benchmark
 from sglang.jit_kernel.per_tensor_quant_fp8 import per_tensor_quant_fp8
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")

 try:
    from vllm import _custom_ops as ops
@@ -20,7 +20,7 @@ from sglang.srt.utils.bench_utils import bench_kineto
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.utils import is_in_ci

-register_cuda_ci(est_time=13, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=13, suite="base-b-kernel-benchmark-1-gpu-large")

 IS_CI = is_in_ci()

@@ -15,7 +15,7 @@ from sglang.jit_kernel.norm import fused_inplace_qknorm
 from sglang.srt.utils import get_current_device_stream_fast
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=10, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-benchmark-1-gpu-large")

 alt_stream = torch.cuda.Stream()

@@ -12,7 +12,7 @@ from sglang.srt.utils import get_current_device_stream_fast
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.utils import is_in_ci

-register_cuda_ci(est_time=12, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=12, suite="base-b-kernel-benchmark-1-gpu-large")

 IS_CI = is_in_ci()

@@ -9,7 +9,7 @@ from sglang.jit_kernel.benchmark.utils import run_benchmark_no_cudagraph
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.utils import is_in_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-benchmark-1-gpu-large")


 def torch_top_k_renorm_probs(probs, top_k):
@@ -13,7 +13,7 @@ from sglang.jit_kernel.resolve_future_token_ids import resolve_future_token_ids_
 from sglang.srt.utils import get_compiler_backend
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=10, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-benchmark-1-gpu-large")
 register_amd_ci(est_time=10, suite="jit-kernel-unit-test-amd")

 SIZE_LIST = get_benchmark_range(
@@ -12,7 +12,7 @@ from sglang.jit_kernel.benchmark.utils import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=6, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=6, suite="base-b-kernel-benchmark-1-gpu-large")

 MAX_SEQ_LEN = 131072
 ROPE_BASE = 10000.0
@@ -26,7 +26,7 @@ from sglang.srt.mem_cache.utils import set_mla_kv_buffer_kernel as sglang_triton
 from sglang.srt.mem_cache.utils import set_mla_kv_buffer_triton as sglang_wrapper
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=9, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=9, suite="base-b-kernel-benchmark-1-gpu-large")


 def _triton_baseline(kv_buffer, loc, cache_k_nope, cache_k_rope):
@@ -14,7 +14,7 @@ from sglang.jit_kernel.benchmark.utils import (
 from sglang.jit_kernel.kvcache import store_cache
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=9, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=9, suite="base-b-kernel-benchmark-1-gpu-large")


 def sglang_jit_store_cache(
@@ -19,7 +19,7 @@ from sglang.test.ci.ci_register import register_cuda_ci

 register_cuda_ci(
    est_time=120,
-    suite="stage-b-kernel-benchmark-1-gpu-large",
+    suite="base-b-kernel-benchmark-1-gpu-large",
    disabled="requires multi-GPU, self-skips in CI",
 )

@@ -17,7 +17,7 @@ from sglang.utils import is_in_ci

 register_cuda_ci(
    est_time=120,
-    suite="stage-b-kernel-benchmark-1-gpu-large",
+    suite="base-b-kernel-benchmark-1-gpu-large",
    disabled="standalone diffusion NVFP4 benchmark",
 )

@@ -20,7 +20,7 @@ from sglang.utils import is_in_ci

 register_cuda_ci(
    est_time=17,
-    suite="stage-b-kernel-benchmark-1-gpu-large",
+    suite="base-b-kernel-benchmark-1-gpu-large",
    disabled="Temporarily skipped to unblock flashinfer upgrade. Ref: https://github.com/sgl-project/sglang/actions/runs/23735552939/job/69139238979?pr=21422",
 )

@@ -16,7 +16,7 @@ from sglang.utils import is_in_ci

 register_cuda_ci(
    est_time=45,
-    suite="stage-b-kernel-benchmark-1-gpu-large",
+    suite="base-b-kernel-benchmark-1-gpu-large",
    disabled="standalone benchmark",
 )

@@ -24,7 +24,7 @@ from sglang.utils import is_in_ci

 register_cuda_ci(
    est_time=120,
-    suite="stage-b-kernel-benchmark-1-gpu-large",
+    suite="base-b-kernel-benchmark-1-gpu-large",
    disabled="self-skips in CI, standalone tool",
 )

@@ -13,7 +13,7 @@ from sglang.jit_kernel.benchmark.utils import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=13, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=13, suite="base-b-kernel-benchmark-1-gpu-large")

 MAX_SEQ_LEN = 131072
 ROPE_BASE = 10000.0
@@ -12,7 +12,7 @@ from sglang.jit_kernel.diffusion.triton.scale_shift import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.utils import is_in_ci

-register_cuda_ci(est_time=13, suite="stage-b-kernel-benchmark-1-gpu-large")
+register_cuda_ci(est_time=13, suite="base-b-kernel-benchmark-1-gpu-large")

 if is_in_ci():
    B_RANGE, S_RANGE, D_RANGE = [1], [128], [3072]
@@ -18,7 +18,7 @@ from sglang.jit_kernel.tests.deepseek_v4.common import (
    to_seq_extend,
 )

-register_cuda_ci(est_time=30, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=30, suite="nightly-kernel-1-gpu", nightly=True)

 Context = Union[LegacyContext, PagedContext]
@@ -18,7 +18,7 @@ from sglang.jit_kernel.tests.deepseek_v4.common import (
    to_seq_extend,
 )

-register_cuda_ci(est_time=30, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=30, suite="nightly-kernel-1-gpu", nightly=True)

 Context = Union[LegacyContext, PagedContext]
@@ -14,7 +14,7 @@ from sglang.srt.layers.quantization.fp8_utils import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=20, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=20, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=80, suite="nightly-kernel-1-gpu", nightly=True)

 DEVICE = "cuda"
@@ -17,7 +17,7 @@ from sglang.srt.layers.quantization.modelopt_quant import pad_nvfp4_weight
 from sglang.test.ci.ci_register import register_cuda_ci

 # B200-only correctness coverage for diffusion NVFP4 scaled mm.
-register_cuda_ci(est_time=15, suite="stage-b-kernel-unit-1-gpu-b200")
+register_cuda_ci(est_time=15, suite="base-b-kernel-unit-1-gpu-b200")

 DEVICE = "cuda"
 DTYPE = torch.bfloat16
@@ -13,7 +13,7 @@ from sglang.jit_kernel.diffusion.cutedsl.scale_residual_norm_scale_shift import
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=28, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=28, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 DEVICE = "cuda"
@@ -9,7 +9,7 @@ from sglang.jit_kernel.diffusion.group_norm_silu import apply_group_norm_silu
 from sglang.jit_kernel.diffusion.triton.group_norm_silu import triton_group_norm_silu
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=8, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=8, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 DEVICE = "cuda"
@@ -8,7 +8,7 @@ import triton
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=44, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=44, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=176, suite="nightly-kernel-1-gpu", nightly=True)

 DEVICE = "cuda"
@@ -12,7 +12,7 @@ from sglang.jit_kernel.diffusion.triton.scale_shift import (
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=15, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=15, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 DEVICE = "cuda"
@@ -8,7 +8,7 @@ from sglang.jit_kernel.activation import SUPPORTED_ACTIVATIONS, run_activation
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=20, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=20, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=30, suite="nightly-kernel-1-gpu", nightly=True)


@@ -6,7 +6,7 @@ import torch
 from sglang.jit_kernel.add_constant import add_constant
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=45, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=45, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=180, suite="nightly-kernel-1-gpu", nightly=True)


@@ -7,7 +7,7 @@ import torch
 from sglang.jit_kernel.awq_dequantize import awq_dequantize as jit_awq_dequantize
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=9, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=9, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 try:
@@ -11,7 +11,7 @@ from sglang.jit_kernel.awq_marlin_repack import (
 from sglang.srt.layers.quantization.utils import pack_cols, quantize_weights
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=10, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -12,7 +12,7 @@ from sglang.srt.layers.quantization.utils import pack_cols, quantize_weights
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import get_weight_perm, marlin_weights

-register_cuda_ci(est_time=10, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -6,7 +6,7 @@ import torch
 from sglang.jit_kernel.clamp_position import clamp_position_cuda
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=12, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=12, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -7,7 +7,7 @@ import triton

 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=17, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=17, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -38,7 +38,7 @@ from sglang.test.ci.ci_register import register_cuda_ci

 register_cuda_ci(
    est_time=300,
-    suite="stage-b-kernel-unit-8-gpu-h200",
+    suite="base-b-kernel-unit-8-gpu-h200",
 )
 register_cuda_ci(
    est_time=300,
@@ -29,7 +29,7 @@ try:
 except ImportError:
    TRITON_AVAILABLE = False

-register_cuda_ci(est_time=5, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -14,7 +14,7 @@ from einops import rearrange, repeat
 from sglang.jit_kernel.flash_attention import flash_attn_varlen_func
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=120, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=120, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=900, suite="nightly-kernel-1-gpu", nightly=True)

 # Skip this test on Hopper machine
@@ -7,7 +7,7 @@ import torch
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -16,7 +16,7 @@ import torch

 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=100, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=100, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=400, suite="nightly-kernel-1-gpu", nightly=True)

 # =============================================================================
@@ -48,7 +48,7 @@ try:
 except ImportError:
    _is_fp8_fnuz = False

-register_cuda_ci(est_time=24, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=24, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 PAGE_SIZE = 64
@@ -26,7 +26,7 @@ try:
 except ImportError:
    KERNELS_AVAILABLE = False

-register_cuda_ci(est_time=6, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=6, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -9,7 +9,7 @@ from sglang.srt.layers.quantization.marlin_utils import marlin_make_workspace
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import awq_marlin_quantize, marlin_quantize

-register_cuda_ci(est_time=13, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=13, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 MNK_FACTORS = [
@@ -13,7 +13,7 @@ from sglang.srt.layers.quantization.utils import (
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import get_weight_perm, marlin_weights

-register_cuda_ci(est_time=16, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=16, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 MARLIN_K_CHUNKS = [128]
@@ -9,7 +9,7 @@ from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.srt.layers.moe.topk import biased_grouped_topk_impl
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=30, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -16,7 +16,7 @@ from sglang.jit_kernel.hadamard import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=128, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=128, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=512, suite="nightly-kernel-1-gpu", nightly=True)

 # Exact M×N Hadamard matrices (±1 entries) copied from
@@ -13,7 +13,7 @@ from sglang.srt.mem_cache.memory_pool_host import (
 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, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 pytestmark = pytest.mark.skipif(
@@ -7,7 +7,7 @@ from sglang.jit_kernel.hisparse import load_cache_to_device_buffer_mla
 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, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 pytestmark = pytest.mark.skipif(
@@ -7,7 +7,7 @@ from sglang.jit_kernel.mla_kv_pack_quantize_fp8 import mla_kv_pack_quantize_fp8
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=60, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=60, suite="base-b-kernel-unit-1-gpu-large")

 DEVICE = "cuda"

@@ -9,7 +9,7 @@ import triton.language as tl
 from sglang.jit_kernel.moe_align import moe_align_block_size
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=28, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=28, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -11,7 +11,7 @@ import torch
 from sglang.jit_kernel.moe_lora_align import moe_lora_align_block_size
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=28, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=28, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -10,7 +10,7 @@ from sglang.srt.layers.moe.fused_moe_triton import moe_align_block_size
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_marlin_utils import awq_marlin_quantize, marlin_quantize

-register_cuda_ci(est_time=10, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -10,7 +10,7 @@ from sglang.jit_kernel.mxfp8 import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -10,7 +10,7 @@ from sglang.jit_kernel.nvfp4 import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 FLOAT4_E2M1_MAX = 6.0
@@ -6,7 +6,7 @@ import torch
 from sglang.jit_kernel.nvfp4 import cutlass_scaled_fp4_mm, scaled_fp4_quant
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -16,7 +16,7 @@ except Exception:

 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=5, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=5, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -8,7 +8,7 @@ import torch
 from sglang.jit_kernel.per_tensor_quant_fp8 import per_tensor_quant_fp8
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=16, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=16, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 try:
@@ -25,7 +25,7 @@ from sglang.srt.layers.quantization.fp8_kernel import (
 )
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=16, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=16, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 configs = list(
@@ -10,7 +10,7 @@ import triton.language as tl
 from sglang.jit_kernel.rope import rotary_embedding
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=18, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=18, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -8,7 +8,7 @@ import triton
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=37, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=37, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=148, suite="nightly-kernel-1-gpu", nightly=True)


@@ -8,7 +8,7 @@ import triton
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=15, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=15, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -9,7 +9,7 @@ import torch

 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=6, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=6, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -6,7 +6,7 @@ import torch
 from sglang.jit_kernel.resolve_future_token_ids import resolve_future_token_ids_cuda
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=9, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=9, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)


@@ -7,7 +7,7 @@ import torch
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=45, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=45, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=240, suite="nightly-kernel-1-gpu", nightly=True)


@@ -13,7 +13,7 @@ from sglang.jit_kernel.rmsnorm_hf import (
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=30, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 EPS = 1e-5
@@ -7,7 +7,7 @@ import triton
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=64, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=64, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=256, suite="nightly-kernel-1-gpu", nightly=True)

 DEVICE = "cuda"
@@ -10,7 +10,7 @@ from sglang.jit_kernel.set_mla_kv_buffer import (
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=30, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=30, suite="base-b-kernel-unit-1-gpu-large")

 DEVICE = "cuda"
 CACHE_SIZE = 4096
@@ -8,7 +8,7 @@ from sglang.jit_kernel.kvcache import can_use_store_cache, store_cache
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci

-register_cuda_ci(est_time=28, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=28, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)
 register_amd_ci(est_time=55, suite="jit-kernel-unit-test-amd")

@@ -16,7 +16,7 @@ from sglang.jit_kernel.timestep_embedding import (
 from sglang.jit_kernel.utils import get_ci_test_range
 from sglang.test.ci.ci_register import register_cuda_ci

-register_cuda_ci(est_time=16, suite="stage-b-kernel-unit-1-gpu-large")
+register_cuda_ci(est_time=16, suite="base-b-kernel-unit-1-gpu-large")
 register_cuda_ci(est_time=120, suite="nightly-kernel-1-gpu", nightly=True)

 CORRECTNESS_BATCH_SIZES = get_ci_test_range(
@@ -16,7 +16,7 @@ from sglang.test.ci.ci_register import register_cuda_ci

 register_cuda_ci(
    est_time=300,
-    suite="stage-b-kernel-unit-8-gpu-h200",
+    suite="base-b-kernel-unit-8-gpu-h200",
 )
 register_cuda_ci(
    est_time=300,