Researcher
Researcher who contributed to SMELT paper on looped transformers cutting training compute.
How media typically covers Wenhao Huang
Wenhao Huang as author
DuPO, a dual preference optimization framework, enables annotation-free LLM feedback via self-supervised reconstruction, achieving 2.13 COMET improvement in translation and 6.4-point gains in mathematical reasoning without costly labels.
“Listed as author of the DuPO paper on preference optimization”
Referenced in coverage
“Researcher who contributed to SMELT paper on looped transformers cutting training compute.”