collaborators
Showing cs.LGShow all

5 papers · 1 filter

cs.LG2026

Access Sets Matter: Budgeting Expert Reads for Scalable Weight-Space Model Merging

Yuanyi Wang, Yanggan Gu, Su Lu +5

Weight-space model merging is usually formulated as an algebraic operation on checkpoints, yet at LLM scale the limiting resource is often the set of expert weights that must be re…

cs.LG2026

Not All Disagreement Is Learnable: Token Teachability in On-Policy Distillation

Yuanyi Wang, Su Lu, Yanggan Gu +6

On-policy distillation (OPD) trains a student on its own rollouts with token-level teacher supervision. Recent selective OPD methods exploit the non-uniformity of OPD signals by pr…

cs.LG2026

OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyond

Zunhai Su, Rui Yang, Chao Zhang +11

The rapid advancement toward long-context reasoning and multi-modal intelligence has made the memory footprint of the Key-Value (KV) cache a dominant memory bottleneck for efficien…

cs.LG2026

FeatCal: Feature Calibration for Post-Merging Models

Yanggan Gu, Shuo Cai, Zihao Wang +7

Model merging combines task experts into one model and avoids joint training, retraining, or deploying many expert models, but the merged model often still underperforms task exper…

cs.LG2026

Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training

Yuanyi Wang, Yifan Yang, Su Lu +9

Continual post-training aims to extend large language models (LLMs) with new knowledge, skills, and behaviors, yet it remains unclear when sequential updates enable capability tran…