Re-enable Flashinfer TRTLLM GEN MHA and Add Unit Test (#12885)

python/sglang/srt/layers/attention/attention_registry.py

@@ -199,7 +199,8 @@ def attn_backend_wrapper(runner: "ModelRunner", full_attn_backend: "AttentionBac
             if is_blackwell():
                 assert (
                     runner.server_args.attention_backend == "triton"
-                ), "triton backend is the only supported backend on Blackwell GPUs for hybrid GDN models, use --attention-backend triton to specify the backend."
+                    or runner.server_args.attention_backend == "trtllm_mha"
+                ), "triton or trtllm_mha backend are the only supported backends on Blackwell GPUs for hybrid GDN models, use --attention-backend triton or --attention-backend trtllm_mha to specify the backend."
             if is_npu():
                 assert (
                     runner.server_args.attention_backend == "ascend"

test/srt/nightly/test_flashinfer_trtllm_gen_attn_backend.py

@@ -0,0 +1,62 @@
+import os
+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 TestFlashinferTrtllmGenAttnBackend(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "Qwen/Qwen3-Next-80B-A3B-Instruct"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            env={**os.environ, "SGLANG_ENABLE_JIT_DEEPGEMM": "False"},
+            other_args=[
+                "--attention-backend",
+                "trtllm_mha",
+                "--cuda-graph-max-bs",
+                "512",
+                "--tp-size",
+                "4",
+                "--ep-size",
+                "4",
+                "--mem-fraction-static",
+                "0.7",
+                "--mamba-ssm-dtype",
+                "bfloat16",
+                "--disable-radix-cache",
+            ],
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_gsm8k(self):
+        args = SimpleNamespace(
+            num_shots=5,
+            data_path=None,
+            num_questions=200,
+            max_new_tokens=512,
+            parallel=128,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval(args)
+        print(f"{metrics=}")
+        self.assertGreater(metrics["accuracy"], 0.93)
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/run_suite.py

@@ -221,6 +221,7 @@ suites = {
     "nightly-4-gpu-b200": [
         TestFile("test_fp4_moe.py", 300),
         TestFile("nightly/test_gpt_oss_4gpu_perf.py", 600),
+        TestFile("nightly/test_flashinfer_trtllm_gen_attn_backend.py", 300),
     ],
     "nightly-8-gpu-b200": [],
     "nightly-4-gpu": [],