Migrate performance, accuracy, and quantization tests to CI registry (#17177)

Co-authored-by: Kangyan-Zhou <zky314343421@gmail.com>

test/registered/perf/test_bench_one_batch_1gpu.py

@@ -0,0 +1,39 @@
+import unittest
+
+from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.test_utils import (
+    DEFAULT_MODEL_NAME_FOR_TEST,
+    DEFAULT_SMALL_MODEL_NAME_FOR_TEST,
+    CustomTestCase,
+    is_in_ci,
+    run_bench_offline_throughput,
+    run_bench_one_batch,
+    write_github_step_summary,
+)
+
+register_cuda_ci(est_time=120, suite="stage-b-test-large-1-gpu-performance")
+
+
+class TestBenchOneBatch1GPU(CustomTestCase):
+
+    def test_bs1_small(self):
+        _, output_throughput, _ = run_bench_one_batch(
+            DEFAULT_SMALL_MODEL_NAME_FOR_TEST, ["--cuda-graph-max-bs", "2"]
+        )
+        self.assertGreater(output_throughput, 50)
+
+    def test_bs1_default(self):
+        output_throughput = run_bench_offline_throughput(
+            DEFAULT_MODEL_NAME_FOR_TEST, ["--cuda-graph-max-bs", "2"]
+        )
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_bs1_default (llama-3.1-8b)\n"
+                f"output_throughput: {output_throughput:.2f} token/s\n"
+            )
+            self.assertGreater(output_throughput, 135)
+
+
+if __name__ == "__main__":
+    unittest.main()