Relax flaky B200 GSM8K accuracy thresholds (#20304)

Co-authored-by: Alison Shao <alisonshao@Mac.attlocal.net>
2026-03-11 12:35:21 -07:00
parent d093e70067
commit 7b44bc923e
2 changed files with 5 additions and 5 deletions
@@ -85,7 +85,7 @@ class TestEagleDPAttnServerSmall(CustomTestCase):

    def test_a_gsm8k(self):
        metrics, avg_spec_accept_length = test_gsm8k(self.base_url)
-        self.assertGreater(metrics["accuracy"], 0.64)
+        self.assertGreater(metrics["accuracy"], 0.62)
        self.assertGreater(avg_spec_accept_length, 2.7)