From e6efe10072f86cd9c1d9eb1b36f12eb800274b9a Mon Sep 17 00:00:00 2001 From: Michael <13900043+michaelzhang-ai@users.noreply.github.com> Date: Thu, 25 Jun 2026 13:12:58 -0700 Subject: [PATCH] [AMD] Register 5 JIT kernel unit tests for AMD nightly CI (#29197) --- test/registered/jit/deepseek_v4/test_c128_v2.py | 3 ++- test/registered/jit/deepseek_v4/test_c4_v2.py | 3 ++- test/registered/jit/diffusion/test_group_norm_silu.py | 3 ++- test/registered/jit/diffusion/test_qwen_image_modulation.py | 3 ++- test/registered/jit/diffusion/test_varlen_pack_pad.py | 3 ++- 5 files changed, 10 insertions(+), 5 deletions(-) diff --git a/test/registered/jit/deepseek_v4/test_c128_v2.py b/test/registered/jit/deepseek_v4/test_c128_v2.py index 2493074d6..105097029 100644 --- a/test/registered/jit/deepseek_v4/test_c128_v2.py +++ b/test/registered/jit/deepseek_v4/test_c128_v2.py @@ -16,10 +16,11 @@ from sglang.jit_kernel.tests.deepseek_v4.common import ( make_state_pool, to_seq_extend, ) -from sglang.test.ci.ci_register import register_cuda_ci +from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci 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) +register_amd_ci(est_time=30, suite="nightly-amd-kernel-1-gpu", nightly=True) Context = Union[LegacyContext, PagedContext] diff --git a/test/registered/jit/deepseek_v4/test_c4_v2.py b/test/registered/jit/deepseek_v4/test_c4_v2.py index cf5d99c07..0cdda3387 100644 --- a/test/registered/jit/deepseek_v4/test_c4_v2.py +++ b/test/registered/jit/deepseek_v4/test_c4_v2.py @@ -16,10 +16,11 @@ from sglang.jit_kernel.tests.deepseek_v4.common import ( make_state_pool, to_seq_extend, ) -from sglang.test.ci.ci_register import register_cuda_ci +from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci 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) +register_amd_ci(est_time=30, suite="nightly-amd-kernel-1-gpu", nightly=True) Context = Union[LegacyContext, PagedContext] diff --git a/test/registered/jit/diffusion/test_group_norm_silu.py b/test/registered/jit/diffusion/test_group_norm_silu.py index 63fde24bf..261960406 100644 --- a/test/registered/jit/diffusion/test_group_norm_silu.py +++ b/test/registered/jit/diffusion/test_group_norm_silu.py @@ -7,10 +7,11 @@ import torch.nn.functional as F 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 +from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci 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) +register_amd_ci(est_time=15, suite="nightly-amd-kernel-1-gpu", nightly=True) DEVICE = "cuda" DTYPES = [torch.float16, torch.bfloat16, torch.float32] diff --git a/test/registered/jit/diffusion/test_qwen_image_modulation.py b/test/registered/jit/diffusion/test_qwen_image_modulation.py index 0cede04c9..26c700ab2 100644 --- a/test/registered/jit/diffusion/test_qwen_image_modulation.py +++ b/test/registered/jit/diffusion/test_qwen_image_modulation.py @@ -10,10 +10,11 @@ from sglang.jit_kernel.diffusion.triton.scale_shift import ( fuse_residual_layernorm_scale_shift_gate_select01_kernel, ) from sglang.jit_kernel.utils import get_ci_test_range -from sglang.test.ci.ci_register import register_cuda_ci +from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci 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) +register_amd_ci(est_time=30, suite="nightly-amd-kernel-1-gpu", nightly=True) DEVICE = "cuda" DTYPES = get_ci_test_range( diff --git a/test/registered/jit/diffusion/test_varlen_pack_pad.py b/test/registered/jit/diffusion/test_varlen_pack_pad.py index d25ba55d9..645e7da63 100644 --- a/test/registered/jit/diffusion/test_varlen_pack_pad.py +++ b/test/registered/jit/diffusion/test_varlen_pack_pad.py @@ -13,10 +13,11 @@ from sglang.jit_kernel.diffusion.triton.varlen_pack_pad import ( fused_scatter_to_padded, ) from sglang.jit_kernel.utils import get_ci_test_range -from sglang.test.ci.ci_register import register_cuda_ci +from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci register_cuda_ci(est_time=10, suite="base-b-kernel-unit-1-gpu-large") register_cuda_ci(est_time=60, suite="nightly-kernel-1-gpu", nightly=True) +register_amd_ci(est_time=15, suite="nightly-amd-kernel-1-gpu", nightly=True) DEVICE = "cuda" DTYPES = get_ci_test_range([torch.bfloat16, torch.float16], [torch.bfloat16])