refactor: consolidate is_in_ci (jit_kernel, sgl-kernel benchmarks, tests) (#21009)

python/sglang/jit_kernel/benchmark/bench_awq_dequantize.py

@@ -5,7 +5,8 @@ import triton
 import triton.testing
 
 from sglang.jit_kernel.awq_dequantize import awq_dequantize as jit_awq_dequantize
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
+from sglang.utils import is_in_ci
 
 try:
     from sgl_kernel import awq_dequantize as aot_awq_dequantize

python/sglang/jit_kernel/benchmark/bench_awq_marlin_moe_repack.py

@@ -7,8 +7,9 @@ from sgl_kernel.scalar_type import scalar_types
 from sglang.jit_kernel.awq_marlin_repack import (
     awq_marlin_moe_repack as jit_awq_marlin_moe_repack,
 )
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.srt.layers.quantization.utils import pack_cols, quantize_weights
+from sglang.utils import is_in_ci
 
 AOT_AVAILABLE = hasattr(torch.ops.sgl_kernel, "awq_marlin_moe_repack") and hasattr(
     torch.ops.sgl_kernel.awq_marlin_moe_repack, "default"

python/sglang/jit_kernel/benchmark/bench_awq_marlin_repack.py

@@ -7,8 +7,9 @@ from sgl_kernel.scalar_type import scalar_types
 from sglang.jit_kernel.awq_marlin_repack import (
     awq_marlin_repack as jit_awq_marlin_repack,
 )
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.srt.layers.quantization.utils import pack_cols, quantize_weights
+from sglang.utils import is_in_ci
 
 AOT_AVAILABLE = hasattr(torch.ops.sgl_kernel, "awq_marlin_repack") and hasattr(
     torch.ops.sgl_kernel.awq_marlin_repack, "default"

python/sglang/jit_kernel/benchmark/bench_concat_mla.py

@@ -6,9 +6,10 @@ import triton.testing
 from sgl_kernel import concat_mla_absorb_q as aot_absorb_q
 from sgl_kernel import concat_mla_k as aot_k
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.jit_kernel.concat_mla import concat_mla_absorb_q as jit_absorb_q
 from sglang.jit_kernel.concat_mla import concat_mla_k as jit_k
+from sglang.utils import is_in_ci
 
 IS_CI = is_in_ci()
 

python/sglang/jit_kernel/benchmark/bench_fused_add_rmsnorm.py

@@ -5,8 +5,9 @@ import triton
 import triton.testing
 from flashinfer import fused_add_rmsnorm as fi_fused_add_rmsnorm
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.jit_kernel.norm import fused_add_rmsnorm as jit_fused_add_rmsnorm
+from sglang.utils import is_in_ci
 
 IS_CI = is_in_ci()
 

python/sglang/jit_kernel/benchmark/bench_fused_norm_scale_shift.py

@@ -8,13 +8,14 @@ import torch
 import triton
 import triton.testing
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark_no_cudagraph
+from sglang.jit_kernel.benchmark.utils import run_benchmark_no_cudagraph
 from sglang.multimodal_gen.runtime.layers.layernorm import (
     LayerNormScaleShift,
     RMSNormScaleShift,
     ScaleResidualLayerNormScaleShift,
     ScaleResidualRMSNormScaleShift,
 )
+from sglang.utils import is_in_ci
 
 if is_in_ci():
     B_RANGE, S_RANGE, D_RANGE = [1], [128], [1024]

python/sglang/jit_kernel/benchmark/bench_gptq_marlin.py

@@ -3,10 +3,11 @@ import triton
 import triton.testing
 from sgl_kernel.scalar_type import scalar_types
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.jit_kernel.gptq_marlin import gptq_marlin_gemm as jit_gptq_marlin_gemm
 from sglang.srt.layers.quantization.marlin_utils import marlin_make_workspace
 from sglang.test.test_marlin_utils import marlin_quantize
+from sglang.utils import is_in_ci
 
 AOT_AVAILABLE = hasattr(torch.ops.sgl_kernel, "gptq_marlin_gemm") and hasattr(
     torch.ops.sgl_kernel.gptq_marlin_gemm, "default"

python/sglang/jit_kernel/benchmark/bench_gptq_marlin_repack.py

@@ -3,9 +3,10 @@ import triton
 import triton.testing
 from sgl_kernel.scalar_type import scalar_types
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.jit_kernel.gptq_marlin_repack import gptq_marlin_repack as jit_fn
 from sglang.srt.layers.quantization.utils import gptq_quantize_weights, pack_rows
+from sglang.utils import is_in_ci
 
 AOT_AVAILABLE = hasattr(torch.ops.sgl_kernel, "gptq_marlin_repack") and hasattr(
     torch.ops.sgl_kernel.gptq_marlin_repack, "default"

python/sglang/jit_kernel/benchmark/bench_moe_wna16_marlin.py

@@ -3,10 +3,11 @@ import triton
 import triton.testing
 from sgl_kernel.scalar_type import scalar_types
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.jit_kernel.moe_wna16_marlin import moe_wna16_marlin_gemm as jit_fn
 from sglang.srt.layers.moe.fused_moe_triton import moe_align_block_size
 from sglang.test.test_marlin_utils import marlin_quantize
+from sglang.utils import is_in_ci
 
 AOT_AVAILABLE = hasattr(torch.ops.sgl_kernel, "moe_wna16_marlin_gemm") and hasattr(
     torch.ops.sgl_kernel.moe_wna16_marlin_gemm, "default"

python/sglang/jit_kernel/benchmark/bench_norm.py

@@ -6,9 +6,10 @@ import triton.testing
 from flashinfer.norm import fused_add_rmsnorm as fi_fused_add_rmsnorm
 from flashinfer.norm import rmsnorm as fi_rmsnorm
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 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.utils import is_in_ci
 
 IS_CI = is_in_ci()
 

python/sglang/jit_kernel/benchmark/bench_norm_impls.py

@@ -15,12 +15,13 @@ from typing import Callable
 import torch
 import torch.nn.functional as F
 
-from sglang.jit_kernel.benchmark.utils import DEFAULT_DEVICE, is_in_ci
+from sglang.jit_kernel.benchmark.utils import DEFAULT_DEVICE
 from sglang.jit_kernel.diffusion.triton.norm import norm_infer, rms_norm_fn
 from sglang.jit_kernel.diffusion.triton.rmsnorm_onepass import triton_one_pass_rms_norm
 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.jit_kernel.utils import KERNEL_PATH
+from sglang.utils import is_in_ci
 
 os.environ.setdefault("FLASHINFER_DISABLE_VERSION_CHECK", "1")
 

python/sglang/jit_kernel/benchmark/bench_per_token_group_quant_8bit.py

@@ -5,10 +5,7 @@ import torch
 import triton
 from sgl_kernel.test_utils import create_per_token_group_quant_test_data
 
-from sglang.jit_kernel.benchmark.utils import (
-    get_benchmark_range,
-    is_in_ci,
-)
+from sglang.jit_kernel.benchmark.utils import get_benchmark_range
 from sglang.jit_kernel.per_token_group_quant_8bit import (
     per_token_group_quant_8bit as sglang_per_token_group_quant_8bit,
 )
@@ -20,6 +17,7 @@ from sglang.srt.layers.quantization.fp8_kernel import (
 )
 from sglang.srt.utils import is_hip
 from sglang.srt.utils.bench_utils import bench_kineto
+from sglang.utils import is_in_ci
 
 IS_CI = is_in_ci()
 

python/sglang/jit_kernel/benchmark/bench_qknorm_across_heads.py

@@ -6,9 +6,10 @@ import triton
 import triton.testing
 from sgl_kernel import rmsnorm
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark
+from sglang.jit_kernel.benchmark.utils import run_benchmark
 from sglang.jit_kernel.norm import fused_inplace_qknorm_across_heads
 from sglang.srt.utils import get_current_device_stream_fast
+from sglang.utils import is_in_ci
 
 IS_CI = is_in_ci()
 

python/sglang/jit_kernel/benchmark/bench_qwen_image_modulation.py

@@ -3,13 +3,14 @@ from typing import Tuple
 import torch
 import triton.testing
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark_no_cudagraph
+from sglang.jit_kernel.benchmark.utils import run_benchmark_no_cudagraph
 from sglang.jit_kernel.diffusion.triton.norm import norm_infer
 from sglang.jit_kernel.diffusion.triton.scale_shift import (
     fuse_layernorm_scale_shift_gate_select01_kernel,
     fuse_residual_layernorm_scale_shift_gate_select01_kernel,
     fuse_scale_shift_gate_select01_kernel,
 )
+from sglang.utils import is_in_ci
 
 if is_in_ci():
     B_RANGE, S_RANGE, D_RANGE = [1], [128], [3072]

python/sglang/jit_kernel/benchmark/bench_renorm.py

@@ -5,7 +5,8 @@ import torch
 import triton
 import triton.testing
 
-from sglang.jit_kernel.benchmark.utils import is_in_ci, run_benchmark_no_cudagraph
+from sglang.jit_kernel.benchmark.utils import run_benchmark_no_cudagraph
+from sglang.utils import is_in_ci
 
 
 def torch_top_k_renorm_probs(probs, top_k):

python/sglang/jit_kernel/benchmark/utils.py

@@ -5,7 +5,7 @@ from typing import Callable, List, Tuple
 import torch
 import triton.testing
 
-from sglang.jit_kernel.utils import is_in_ci as jit_kernel_is_in_ci
+from sglang.utils import is_in_ci
 
 # Common constants
 DEFAULT_DTYPE = torch.bfloat16
@@ -13,11 +13,6 @@ DEFAULT_DEVICE = "cuda"
 DEFAULT_QUANTILES = [0.5, 0.2, 0.8]
 
 
-def is_in_ci() -> bool:
-    """Check if running in CI environment."""
-    return jit_kernel_is_in_ci()
-
-
 def get_benchmark_range(full_range: List, ci_range: List) -> List:
     """Return appropriate benchmark range based on CI environment."""
     return ci_range if is_in_ci() else full_range

python/sglang/jit_kernel/utils.py

@@ -7,6 +7,8 @@ from typing import TYPE_CHECKING, Any, Callable, List, Tuple, TypeAlias, TypeVar
 
 import torch
 
+from sglang.utils import is_in_ci
+
 if TYPE_CHECKING:
     from tvm_ffi import Module
 
@@ -14,11 +16,6 @@ F = TypeVar("F", bound=Callable[..., Any])
 _FULL_TEST_ENV_VAR = "SGLANG_JIT_KERNEL_RUN_FULL_TESTS"
 
 
-def is_in_ci() -> bool:
-    ci_env_vars = ("SGLANG_IS_IN_CI", "CI", "GITHUB_ACTIONS")
-    return any(os.getenv(env_var, "false").lower() == "true" for env_var in ci_env_vars)
-
-
 def should_run_full_tests() -> bool:
     return os.getenv(_FULL_TEST_ENV_VAR, "false").lower() == "true"