diff --git a/test/srt/test_nightly_gsm8k_eval.py b/test/srt/test_nightly_gsm8k_eval.py index 35e7d6eb7..08faae3f4 100644 --- a/test/srt/test_nightly_gsm8k_eval.py +++ b/test/srt/test_nightly_gsm8k_eval.py @@ -52,7 +52,7 @@ class TestEvalAccuracyLarge(unittest.TestCase): if is_tp2: other_args.extend(["--tp", "2"]) if "DeepSeek" in model: - other_args.append("--enable-mla") + other_args.extend(["--enable-mla", "--mem-frac", "0.85"]) self.process = popen_launch_server( model,