Improve streaming, log_level, memory report, weight loading, and benchmark script (#7632)

Co-authored-by: Kan Wu <wukanustc@gmail.com>

test/srt/run_suite.py

@@ -173,10 +173,11 @@ suites = {
         # TestFile("test_deepep_intranode.py", 50),
         # TestFile("test_deepep_low_latency.py", 50),
         # TestFile("test_moe_deepep_eval_accuracy_large.py", 250),
+        # Disabled because it hangs on the CI.
+        # TestFile("test_moe_ep.py", 181),
         TestFile("test_disaggregation.py", 270),
         TestFile("test_disaggregation_different_tp.py", 155),
         TestFile("test_full_deepseek_v3.py", 463),
-        TestFile("test_moe_ep.py", 181),
     ],
     "per-commit-8-gpu-amd": [
         TestFile("test_full_deepseek_v3.py", 250),

test/srt/test_vision_chunked_prefill.py

@@ -178,7 +178,7 @@ class TestVisionChunkedPrefill(CustomTestCase):
             print(output_chunked)
             print("output without chunked prefill:")
             print(output_no_chunked)
-            assert output_chunked == output_no_chunked
+            self.assertEqual(output_chunked, output_no_chunked)
 
     def test_chunked_prefill(self):
         self._test_chunked_prefill(batches=[False, True], num_frames=[1, [2, 6, 8, 10]])