[Test] Replace NVFP4 MoE runner backend e2e matrix with a layer-level unit test (#33611)

2026-08-04 16:03:29 -07:00
parent 0d99d91e49
commit 76dc89f5aa
4 changed files with 233 additions and 234 deletions
@@ -1,87 +0,0 @@
-"""Extra: DeepSeek-V3 FP4 with FlashInfer Cutlass MoE backend.
-
-Sibling per-commit file (test_deepseek_v3_fp4_4gpu.py) keeps the
-SymmetricMemory variant.
-"""
-
-import os
-import unittest
-from types import SimpleNamespace
-
-from sglang.srt.utils import kill_process_tree
-from sglang.test.ci.ci_register import register_cuda_ci
-from sglang.test.run_eval import run_eval
-from sglang.test.test_utils import (
-    DEFAULT_URL_FOR_TEST,
-    CustomTestCase,
-    is_in_ci,
-    popen_launch_server,
-    write_github_step_summary,
-)
-
-register_cuda_ci(est_time=960, stage="extra-b", runner_config="4-gpu-b200")
-
-FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
-SERVER_LAUNCH_TIMEOUT = 1200
-
-
-class TestDeepseekV3FP4CutlassMoE(CustomTestCase):
-    @classmethod
-    def setUpClass(cls):
-        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
-        cls.base_url = DEFAULT_URL_FOR_TEST
-        other_args = [
-            "--tp",
-            "4",
-            "--ep",
-            "4",
-            "--attention-backend",
-            "trtllm_mla",
-            "--moe-runner-backend",
-            "flashinfer_cutlass",
-            "--quantization",
-            "modelopt_fp4",
-            "--model-loader-extra-config",
-            '{"enable_multithread_load": true}',
-        ]
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=SERVER_LAUNCH_TIMEOUT,
-            other_args=other_args,
-            env={
-                **os.environ,
-                "SGLANG_MOE_NVFP4_DISPATCH": "1",  # Enable nvfp4 all gather
-            },
-        )
-
-    @classmethod
-    def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
-
-    def test_a_gsm8k(
-        self,
-    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
-        args = SimpleNamespace(
-            base_url=self.base_url,
-            model=self.model,
-            eval_name="gsm8k",
-            api="completion",
-            max_tokens=512,
-            num_examples=1319,
-            num_threads=1319,
-            num_shots=8,
-        )
-        metrics = run_eval(args)
-        print(f"{metrics=}")
-
-        if is_in_ci():
-            write_github_step_summary(
-                f"### test_gsm8k (deepseek-v3-fp4-cutlass-moe)\n"
-                f'{metrics["score"]=:.3f}\n'
-            )
-            self.assertGreater(metrics["score"], 0.93)
-
-
-if __name__ == "__main__":
-    unittest.main()