From b0a25d0913307717eaeb9e036ce8a4e1a26224e3 Mon Sep 17 00:00:00 2001 From: b8zhong Date: Tue, 9 Dec 2025 23:08:41 -0800 Subject: [PATCH] fix b200 ci (#14786) --- test/srt/run_suite.py | 4 ++-- test/srt/test_llama31_fp4.py | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py index 591e3ca60..ebe766925 100644 --- a/test/srt/run_suite.py +++ b/test/srt/run_suite.py @@ -171,8 +171,8 @@ suites = { 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 36ae36971..be1b04648 100644 --- a/test/srt/test_llama31_fp4.py +++ b/test/srt/test_llama31_fp4.py @@ -49,7 +49,7 @@ class TestLlama31FP4(unittest.TestCase): metrics = run_eval_few_shot_gsm8k(args) print(metrics) - self.assertGreater(metrics["accuracy"], 0.54) + self.assertGreater(metrics["accuracy"], 0.64) if __name__ == "__main__":