DiDPO splits code diffs into similar sub-diff anchors across rollouts and uses those groups to compute fine-grained token-level advantages, outperforming GRPO and GiGPO on coding benchmarks.
A general theo- retical paradigm to understand learning from hu- man preferences
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
DiDPO: Diff-in-Diff Policy Optimization for Coding Agent Training
DiDPO splits code diffs into similar sub-diff anchors across rollouts and uses those groups to compute fine-grained token-level advantages, outperforming GRPO and GiGPO on coding benchmarks.