Fix quantization and nightly tests (#4258)

test/srt/run_suite.py

@@ -22,6 +22,7 @@ suites = {
         TestFile("models/test_reward_models.py", 83),
         TestFile("models/test_gme_qwen_models.py", 45),
         TestFile("test_abort.py", 51),
+        TestFile("test_awq.py"),
         TestFile("test_block_int8.py", 22),
         TestFile("test_chunked_prefill.py", 336),
         TestFile("test_eagle_infer.py", 447),

test/srt/test_awq.py

@@ -0,0 +1,44 @@
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.run_eval import run_eval
+from sglang.test.test_utils import (
+    DEFAULT_AWQ_MOE_MODEL_NAME_FOR_TEST,
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    popen_launch_server,
+)
+
+
+class TestAWQ(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = DEFAULT_AWQ_MOE_MODEL_NAME_FOR_TEST
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=["--trust-remote-code"],
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_mmlu(self):
+        args = SimpleNamespace(
+            base_url=self.base_url,
+            model=self.model,
+            eval_name="mmlu",
+            num_examples=64,
+            num_threads=32,
+        )
+
+        metrics = run_eval(args)
+        self.assertGreater(metrics["score"], 0.65)
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/test_nightly_gsm8k_eval.py

@@ -38,6 +38,7 @@ MODEL_SCORE_THRESHOLDS = {
     "neuralmagic/Qwen2-57B-A14B-Instruct-FP8": 0.82,
     "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4": 0.84,
     "hugging-quants/Meta-Llama-3.1-8B-Instruct-GPTQ-INT4": 0.83,
+    "hugging-quants/Mixtral-8x7B-Instruct-v0.1-AWQ-INT4": 0.60,
 }