ci: lower VL PP gsm8k threshold to 0.60 (0.65 has zero margin on H100) (#31702)

2026-07-19 15:41:12 -07:00
parent 7d64858093
commit 1a317839d7
1 changed files with 1 additions and 1 deletions
@@ -66,7 +66,7 @@ class TestQwenVLPPAccuracy(unittest.TestCase):
        metrics = run_eval(args)
        print(f"{metrics=}")

-        self.assertGreaterEqual(metrics["score"], 0.65)
+        self.assertGreaterEqual(metrics["score"], 0.60)
        # Wait a little bit so that the memory check happens.
        time.sleep(4)