diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py index 591e3ca604a0..ebe766925d21 100644 --- a/test/srt/run_suite.py +++ b/test/srt/run_suite.py @@ -171,8 +171,8 @@ TestFile("test_deepseek_v3_fp4_4gpu.py", 1800), TestFile("test_flash_attention_4.py", 300), TestFile("test_fp8_blockwise_gemm.py", 280), - TestFile("test_gpt_oss_4gpu.py", 600), - TestFile("test_llama31_fp4.py", 300), + TestFile("test_gpt_oss_4gpu.py", 700), + TestFile("test_llama31_fp4.py", 90), TestFile("test_eagle_infer_beta_dp_attention.py", 300), ], # "per-commit-8-gpu-b200": [ diff --git a/test/srt/test_llama31_fp4.py b/test/srt/test_llama31_fp4.py index 36ae3697114f..be1b04648420 100644 --- a/test/srt/test_llama31_fp4.py +++ b/test/srt/test_llama31_fp4.py @@ -49,7 +49,7 @@ def test_gsm8k(self): metrics = run_eval_few_shot_gsm8k(args) print(metrics) - self.assertGreater(metrics["accuracy"], 0.54) + self.assertGreater(metrics["accuracy"], 0.64) if __name__ == "__main__":