From ff0f370f85319ca57f6a1572da0c1dfe22761cb3 Mon Sep 17 00:00:00 2001 From: Alison Shao <54658187+alisonshao@users.noreply.github.com> Date: Sun, 4 Jan 2026 11:51:12 -0800 Subject: [PATCH] ci: migrate MoE tests to test/registered/moe/ (#16127) --- test/{srt => registered/moe}/test_cutedsl_moe.py | 4 ++++ test/{srt => registered/moe}/test_fused_moe.py | 5 +++++ .../models => registered/moe}/test_glm4_moe_models.py | 4 ++++ test/{srt/ep => registered/moe}/test_moe_ep.py | 4 ++++ test/{srt => registered/moe}/test_torch_compile_moe.py | 4 ++++ test/{srt => registered/moe}/test_triton_fused_moe.py | 4 ++++ .../moe}/test_triton_moe_channel_fp8_kernel.py | 4 ++++ test/srt/run_suite.py | 8 -------- 8 files changed, 29 insertions(+), 8 deletions(-) rename test/{srt => registered/moe}/test_cutedsl_moe.py (99%) rename test/{srt => registered/moe}/test_fused_moe.py (97%) rename test/{srt/models => registered/moe}/test_glm4_moe_models.py (91%) rename test/{srt/ep => registered/moe}/test_moe_ep.py (94%) rename test/{srt => registered/moe}/test_torch_compile_moe.py (94%) rename test/{srt => registered/moe}/test_triton_fused_moe.py (98%) rename test/{srt => registered/moe}/test_triton_moe_channel_fp8_kernel.py (98%) diff --git a/test/srt/test_cutedsl_moe.py b/test/registered/moe/test_cutedsl_moe.py similarity index 99% rename from test/srt/test_cutedsl_moe.py rename to test/registered/moe/test_cutedsl_moe.py index d60ce723f..87a133947 100644 --- a/test/srt/test_cutedsl_moe.py +++ b/test/registered/moe/test_cutedsl_moe.py @@ -1,4 +1,8 @@ # SPDX-License-Identifier: Apache-2.0 +from sglang.test.ci.ci_register import register_cuda_ci + +register_cuda_ci(est_time=300, suite="stage-c-test-large-4-gpu-b200") + import unittest from typing import Callable diff --git a/test/srt/test_fused_moe.py b/test/registered/moe/test_fused_moe.py similarity index 97% rename from test/srt/test_fused_moe.py rename to test/registered/moe/test_fused_moe.py index 65d35c59a..96a2418f7 100644 --- a/test/srt/test_fused_moe.py +++ b/test/registered/moe/test_fused_moe.py @@ -1,3 +1,8 @@ +from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci + +register_cuda_ci(est_time=80, suite="stage-b-test-small-1-gpu") +register_amd_ci(est_time=30, suite="stage-a-test-1") + import unittest import torch diff --git a/test/srt/models/test_glm4_moe_models.py b/test/registered/moe/test_glm4_moe_models.py similarity index 91% rename from test/srt/models/test_glm4_moe_models.py rename to test/registered/moe/test_glm4_moe_models.py index cb742c699..5ac178f7d 100644 --- a/test/srt/models/test_glm4_moe_models.py +++ b/test/registered/moe/test_glm4_moe_models.py @@ -1,3 +1,7 @@ +from sglang.test.ci.ci_register import register_cuda_ci + +register_cuda_ci(est_time=100, suite="stage-b-test-large-2-gpu") + import unittest from types import SimpleNamespace diff --git a/test/srt/ep/test_moe_ep.py b/test/registered/moe/test_moe_ep.py similarity index 94% rename from test/srt/ep/test_moe_ep.py rename to test/registered/moe/test_moe_ep.py index 74a5790d4..d2fe766b4 100644 --- a/test/srt/ep/test_moe_ep.py +++ b/test/registered/moe/test_moe_ep.py @@ -1,3 +1,7 @@ +from sglang.test.ci.ci_register import register_cuda_ci + +register_cuda_ci(est_time=140, suite="stage-b-test-large-2-gpu") + import unittest from types import SimpleNamespace diff --git a/test/srt/test_torch_compile_moe.py b/test/registered/moe/test_torch_compile_moe.py similarity index 94% rename from test/srt/test_torch_compile_moe.py rename to test/registered/moe/test_torch_compile_moe.py index 8bc7b45d3..bcfac3270 100644 --- a/test/srt/test_torch_compile_moe.py +++ b/test/registered/moe/test_torch_compile_moe.py @@ -1,3 +1,7 @@ +from sglang.test.ci.ci_register import register_cuda_ci + +register_cuda_ci(est_time=210, suite="stage-b-test-small-1-gpu") + import time import unittest from types import SimpleNamespace diff --git a/test/srt/test_triton_fused_moe.py b/test/registered/moe/test_triton_fused_moe.py similarity index 98% rename from test/srt/test_triton_fused_moe.py rename to test/registered/moe/test_triton_fused_moe.py index 07e774a9b..97ae690da 100644 --- a/test/srt/test_triton_fused_moe.py +++ b/test/registered/moe/test_triton_fused_moe.py @@ -1,3 +1,7 @@ +from sglang.test.ci.ci_register import register_cuda_ci + +register_cuda_ci(est_time=12, suite="stage-b-test-small-1-gpu") + import unittest import torch diff --git a/test/srt/test_triton_moe_channel_fp8_kernel.py b/test/registered/moe/test_triton_moe_channel_fp8_kernel.py similarity index 98% rename from test/srt/test_triton_moe_channel_fp8_kernel.py rename to test/registered/moe/test_triton_moe_channel_fp8_kernel.py index ab2809257..82a983260 100644 --- a/test/srt/test_triton_moe_channel_fp8_kernel.py +++ b/test/registered/moe/test_triton_moe_channel_fp8_kernel.py @@ -1,3 +1,7 @@ +from sglang.test.ci.ci_register import register_cuda_ci + +register_cuda_ci(est_time=16, suite="stage-b-test-small-1-gpu") + import itertools import unittest diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py index 52be01790..0ebac3879 100644 --- a/test/srt/run_suite.py +++ b/test/srt/run_suite.py @@ -54,7 +54,6 @@ suites = { TestFile("test_external_models.py", 30), TestFile("test_fp8_utils.py", 9), TestFile("rotary_embedding/test_mrope.py", 10), - TestFile("test_fused_moe.py", 80), TestFile("test_gpt_oss_1gpu.py", 402), TestFile("test_harmony_parser.py", 6), TestFile("test_hidden_states.py", 55), @@ -88,10 +87,7 @@ suites = { TestFile("test_profile_merger_http_api.py", 9), TestFile("test_swa_unittest.py", 8), TestFile("test_torch_compile.py", 190), - TestFile("test_torch_compile_moe.py", 210), - TestFile("test_triton_fused_moe.py", 12), TestFile("test_torchao.py", 103), - TestFile("test_triton_moe_channel_fp8_kernel.py", 16), TestFile("test_utils_update_weights.py", 29), TestFile("test_video_utils.py", 5), TestFile("test_vision_chunked_prefill.py", 150), @@ -100,12 +96,10 @@ suites = { TestFile("test_modelopt_export.py", 9), ], "per-commit-2-gpu": [ - TestFile("ep/test_moe_ep.py", 140), TestFile("hicache/test_hicache_storage_3fs_backend.py", 200), TestFile("hicache/test_hicache_storage_file_backend.py", 200), TestFile("hicache/test_hicache_storage_mooncake_backend.py", 300), TestFile("layers/attention/mamba/test_mamba2_mixer.py", 50), - TestFile("models/test_glm4_moe_models.py", 100), TestFile("models/test_kimi_linear_models.py", 90), TestFile("models/test_nvidia_nemotron_nano_v2.py", 132), TestFile("rl/test_update_weights_from_distributed.py", 103), @@ -149,7 +143,6 @@ suites = { # TestFile("test_mistral_large3_basic.py", 275), # Moved to nightly - large model # ], "per-commit-4-gpu-gb200": [ - TestFile("test_cutedsl_moe.py", 300), TestFile("test_deepseek_v3_cutedsl_4gpu.py", 1800), ], "per-commit-4-gpu-deepep": [ @@ -233,7 +226,6 @@ suite_amd = { TestFile("test_constrained_decoding.py", 120), TestFile("test_eval_fp8_accuracy.py", 303), TestFile("test_external_models.py", 45), - TestFile("test_fused_moe.py", 30), TestFile("test_harmony_parser.py", 20), TestFile("test_input_embeddings.py", 38), TestFile("test_io_struct.py", 8),