ci: prune per-commit CUDA tests — move 25 files + 13 testcases to test/manual/ (#24721)

test/manual/scheduler/test_no_chunked_prefill.py

@@ -0,0 +1,34 @@
+import unittest
+
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.test_utils import (
+    DEFAULT_MODEL_NAME_FOR_TEST,
+    CustomTestCase,
+    run_bench_serving,
+    run_mmlu_test,
+)
+
+register_cuda_ci(est_time=131, suite="stage-b-test-1-gpu-large")
+register_amd_ci(est_time=108, suite="stage-b-test-1-gpu-small-amd")
+
+
+class TestNoChunkedPrefill(CustomTestCase):
+
+    def test_no_chunked_prefill(self):
+        run_mmlu_test(
+            disable_radix_cache=False, enable_mixed_chunk=False, chunked_prefill_size=-1
+        )
+
+    def test_no_chunked_prefill_without_radix_cache(self):
+        res = run_bench_serving(
+            model=DEFAULT_MODEL_NAME_FOR_TEST,
+            num_prompts=10,
+            request_rate=float("inf"),
+            other_server_args=["--disable-radix-cache", "--chunked-prefill-size", "-1"],
+        )
+
+        assert res["completed"] == 10
+
+
+if __name__ == "__main__":
+    unittest.main()

test/manual/scheduler/test_no_overlap_scheduler.py

@@ -0,0 +1,39 @@
+"""
+Usage:
+python3 -m unittest test_overlap_schedule.TestOverlapSchedule.test_radix_attention_chunked_prefill
+python3 test_overlap_schedule.py
+"""
+
+import unittest
+
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.test_utils import CustomTestCase, run_mmlu_test
+
+register_cuda_ci(est_time=267, suite="stage-b-test-1-gpu-large")
+register_amd_ci(est_time=275, suite="stage-b-test-1-gpu-small-amd")
+
+
+class TestOverlapSchedule(CustomTestCase):
+    def test_no_radix_attention_chunked_prefill(self):
+        run_mmlu_test(
+            disable_radix_cache=True, chunked_prefill_size=32, disable_overlap=True
+        )
+
+    def test_no_radix_attention_no_chunked_prefill(self):
+        run_mmlu_test(
+            disable_radix_cache=True, chunked_prefill_size=-1, disable_overlap=True
+        )
+
+    def test_radix_attention_chunked_prefill(self):
+        run_mmlu_test(
+            disable_radix_cache=False, chunked_prefill_size=32, disable_overlap=True
+        )
+
+    def test_radix_attention_no_chunked_prefill(self):
+        run_mmlu_test(
+            disable_radix_cache=False, chunked_prefill_size=-1, disable_overlap=True
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()