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

2026-05-08 15:53:23 -07:00
parent aefd8e257f
commit 5fbec0e445
45 changed files with 818 additions and 525 deletions
@@ -0,0 +1,113 @@
+"""Archived test classes split out of test/registered/mla/test_mla_int8_deepseek_v3.py.
+
+Originally registered with `register_cuda_ci(...)`. Moved here as part of
+the per-commit pruning effort to keep the code reachable manually.
+Run with `python3 test/manual/mla/test_mla_int8_deepseek_v3_archived.py`.
+"""
+
+import unittest
+from types import SimpleNamespace
+
+import torch
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.run_eval import run_eval
+from sglang.test.test_utils import (
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    is_in_ci,
+    popen_launch_server,
+)
+
+
+# DeepSeek-V3 INT8 quantization tests (channel and block INT8)
+class TestMLADeepseekV3ChannelInt8(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "lmsys/sglang-ci-dsv3-channel-int8-test"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = ["--trust-remote-code"]
+        if torch.cuda.is_available() and torch.version.cuda:
+            other_args.extend(
+                [
+                    "--cuda-graph-max-bs",
+                    "16",
+                    "--enable-torch-compile",
+                    "--torch-compile-max-bs",
+                    "2",
+                ]
+            )
+        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(
+            base_url=self.base_url,
+            model=self.model,
+            eval_name="gsm8k",
+            api="completion",
+            max_tokens=512,
+            num_examples=200,
+            num_threads=128,
+        )
+        metrics = run_eval(args)
+        print(metrics)
+
+        self.assertGreaterEqual(metrics["score"], 0.61)
+
+
+@unittest.skipIf(is_in_ci(), "To reduce the CI execution time.")
+class TestMLADeepseekV3BlockInt8(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "lmsys/sglang-ci-dsv3-block-int8-test"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = ["--trust-remote-code"]
+        if torch.cuda.is_available() and torch.version.cuda:
+            other_args.extend(
+                [
+                    "--cuda-graph-max-bs",
+                    "16",
+                    "--enable-torch-compile",
+                    "--torch-compile-max-bs",
+                    "2",
+                ]
+            )
+        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(
+            base_url=self.base_url,
+            model=self.model,
+            eval_name="gsm8k",
+            api="completion",
+            max_tokens=512,
+            num_examples=200,
+            num_threads=128,
+        )
+        metrics = run_eval(args)
+        print(metrics)
+
+        self.assertGreater(metrics["score"], 0.62)
+
+
+if __name__ == "__main__":
+    unittest.main()