[CI] Fix flaky pp single node test (#6689)

Signed-off-by: Shangming Cai <caishangming@linux.alibaba.com>
This commit is contained in:
shangmingc
2025-05-28 15:40:26 +08:00
committed by GitHub
parent fba03b29e3
commit c25231c679

View File

@@ -108,11 +108,13 @@ class TestQwenPPAccuracy(unittest.TestCase):
print(f"[Qwen PP Comparison] Baseline: {baseline} | PP: {pp_metrics}")
self.assertAlmostEqual(
self.assertGreaterEqual(
pp_metrics["accuracy"],
baseline["accuracy"],
delta=0.01,
msg=f"PP accuracy exceeds 1% (baseline: {baseline['accuracy']}, pp: {pp_metrics['accuracy']})",
baseline["accuracy"] - 0.01,
msg=(
f"PP accuracy dropped more than 1% compared to baseline. "
f"Baseline: {baseline['accuracy']:.2%}, PP: {pp_metrics['accuracy']:.2%}"
),
)
@@ -164,11 +166,13 @@ class TestQwenPPTieWeightsAccuracy(unittest.TestCase):
print(f"[Qwen PP Comparison] Baseline: {baseline} | PP: {pp_metrics}")
self.assertAlmostEqual(
self.assertGreaterEqual(
pp_metrics["accuracy"],
baseline["accuracy"],
delta=0.01,
msg=f"PP accuracy exceeds 1% (baseline: {baseline['accuracy']}, pp: {pp_metrics['accuracy']})",
baseline["accuracy"] - 0.01,
msg=(
f"PP accuracy dropped more than 1% compared to baseline. "
f"Baseline: {baseline['accuracy']:.2%}, PP: {pp_metrics['accuracy']:.2%}"
),
)