[Feat][NVFP4] Enable NVFP4 MoE for Qwen series models (eg. Qwen3-Next) #13761 (#13761)

Co-authored-by: Kaixi Hou <kaixih@nvidia.com>

test/nightly/test_qwen3_fp4_trtllm_gen_moe.py

@@ -0,0 +1,61 @@
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.few_shot_gsm8k import run_eval
+from sglang.test.test_utils import (
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+
+class TestFlashinferTrtllmGenMoeBackend(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "nvidia/Qwen3-30B-A3B-NVFP4"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=[
+                "--moe-runner-backend",
+                "flashinfer_trtllm",
+                "--quantization",
+                "modelopt_fp4",
+                "--trust-remote-code",
+                "--disable-radix-cache",
+                "--max-running-requests",
+                "1024",
+                "--chunked-prefill-size",
+                "16384",
+                "--mem-fraction-static",
+                "0.89",
+                "--max-prefill-tokens",
+                "16384",
+            ],
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_gsm8k(self):
+        args = SimpleNamespace(
+            num_shots=8,
+            data_path=None,
+            num_questions=1319,
+            max_new_tokens=512,
+            parallel=1319,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval(args)
+        print(f"{metrics=}")
+        self.assertGreater(metrics["accuracy"], 0.88)
+
+
+if __name__ == "__main__":
+    unittest.main()

test/run_suite_nightly.py

@@ -24,6 +24,7 @@ suites = {
         TestFile("test_flashinfer_trtllm_gen_attn_backend.py", 300),
         TestFile("test_deepseek_v3_fp4_cutlass_moe.py", 900),
         TestFile("test_fp4_moe.py", 300),
+        TestFile("test_qwen3_fp4_trtllm_gen_moe.py", 300),
     ],
     "nightly-8-gpu-b200": [
         TestFile("test_deepseek_r1_fp8_trtllm_backend.py", 3600),