diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index 13cd1533a..e2026160f 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -1517,11 +1517,6 @@ class ServerArgs: self.ep_size == 1 ), "FP8 Cutlass MoE is only supported with ep_size == 1" - if self.moe_runner_backend == "deep_gemm": - assert ( - self.ep_size > 1 and self.moe_a2a_backend == "deepep" - ), "DeepGemm MoE runner is only supported when ep is enabled and moe_a2a_backend is deepep" - def _handle_a2a_moe(self): if self.moe_a2a_backend == "deepep": if self.deepep_mode == "normal": diff --git a/test/srt/ep/test_moe_ep.py b/test/srt/ep/test_moe_ep.py index bccd2673a..74a5790d4 100644 --- a/test/srt/ep/test_moe_ep.py +++ b/test/srt/ep/test_moe_ep.py @@ -66,8 +66,6 @@ class TestEpDeepGEMM(CustomTestCase): "fp8", "--moe-runner-backend", "deep_gemm", - "--moe-a2a-backend", - "deepep", ], ) diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py index 6574ac8a6..13f880d53 100644 --- a/test/srt/run_suite.py +++ b/test/srt/run_suite.py @@ -135,6 +135,7 @@ suites = { TestFile("test_modelopt_export.py", 30), ], "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), @@ -201,9 +202,6 @@ suites = { ], # Nightly test suites have been moved to test/run_suite_nightly.py "__not_in_ci__": [ - TestFile( - "ep/test_moe_ep.py", 140 - ), # Temporarily disabled, need to fix ep errors TestFile("test_release_memory_occupation.py", 200), # Temporarily disabled TestFile("models/test_dummy_grok_models.py"), TestFile(