[AMD][CI] Fix VLM MMMU nightly max_tokens for CoT prompt (#41979)

This commit is contained in:
Michael
2026-10-02 18:59:05 +08:00
committed by GitHub
parent 895525f062
commit 544e46c080
@@ -229,7 +229,10 @@ class TestNightlyVLMMmmuEvalAMD(unittest.TestCase):
eval_name="mmmu",
num_examples=100,
num_threads=64,
max_tokens=30,
# The MMMU prompt asks for step-by-step reasoning ending in
# "Answer: X" (#21841); 30 tokens truncated it before the
# answer, so the parser fell back to "A" for every model.
max_tokens=1024,
)
# Run evaluation with timing