Add SDAR model support (#18318)

Co-authored-by: root <root@gpu-lg-cmc-h-h200-3047.host.h.pjlab.org.cn> Co-authored-by: chengshuang <chengshuang@pjlab.org.cn> Co-authored-by: 赵晨阳 <zhaochen20@outlook.com>
2026-02-19 11:20:32 -08:00
parent 3207427d6d
commit 44ab752b7a
5 changed files with 1439 additions and 6 deletions
@@ -0,0 +1,91 @@
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+
+register_cuda_ci(est_time=181, suite="stage-b-test-large-1-gpu")
+register_amd_ci(est_time=330, suite="stage-b-test-small-1-gpu-amd")
+
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
+from sglang.test.send_one import BenchArgs, send_one_prompt
+from sglang.test.test_utils import (
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    is_in_amd_ci,
+    is_in_ci,
+    popen_launch_server,
+    write_github_step_summary,
+)
+
+
+class TestSDARMini(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "JetLM/SDAR-8B-Chat"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+
+        other_args = [
+            "--trust-remote-code",
+            "--mem-fraction-static",
+            "0.8",
+            "--max-running-requests",
+            "64",
+            "--attention-backend",
+            "flashinfer",
+            "--dllm-algorithm",
+            "LowConfidence",
+            "--tp",
+            "1",
+        ]
+
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=other_args,
+        )
+
+    @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=1024,
+            parallel=128,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval_few_shot_gsm8k(args)
+        print(f"{metrics=}")
+
+        self.assertGreater(metrics["accuracy"], 0.88)
+        if is_in_amd_ci():
+            self.assertGreater(metrics["output_throughput"], 80)
+        else:
+            self.assertGreater(metrics["output_throughput"], 250)
+
+    def test_bs_1_speed(self):
+        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
+        acc_length, speed = send_one_prompt(args)
+
+        print(f"{speed=:.2f}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_bs_1_speed (SDAR-8B-Chat) with tp1\n"
+                f"{speed=:.2f} token/s\n"
+            )
+            if is_in_amd_ci():
+                self.assertGreater(speed, 10)
+            else:
+                self.assertGreater(speed, 70)
+
+
+if __name__ == "__main__":
+    unittest.main()