refactor qwen moe code, use communicator to support tp+dp (#6581)

2025-05-26 14:01:10 +08:00
parent fc419b62e8
commit 65f091310c
5 changed files with 79 additions and 380 deletions
--- a/test/srt/test_disaggregation.py
+++ b/test/srt/test_disaggregation.py
@@ -146,7 +146,7 @@ class TestDisaggregationAccuracy(CustomTestCase):
        self.assertGreater(metrics["accuracy"], 0.62)

    def test_logprob(self):
-        prompt = "The capital of taiwan is "
+        prompt = "The capital of france is "
        response = requests.post(
            self.lb_url + "/generate",
            json={