Step-2 SFT reference policies (π_ref) used to initialize DDRO (MS MARCO / NQ; PQ and Title+URL DocIDs); use these for fair comparisons/ablations.
Kidist Amde Mekonnen
kiyam
AI & ML interests
AI (Generative models, Computer vision, NLP) ,XAI
Recent Activity
authored a paper about 22 hours ago
Lightweight and Direct Document Relevance Optimization for Generative
Information Retrieval upvoted a collection 10 days ago
Splade-Code liked a model 10 days ago
naver/splade-code-06BOrganizations
None yet