LLM Digest

Story

arxiv_cs_lg · Nov 12, 2026 · paper

Source brief

Joint Null-Space Projection for Data-Free Model Merging

arxiv.orgNov 12, 2026
original source linked

In brief

Merging independently fine-tuned checkpoints into a single multi-task model requires no data, no labels, and no gradient steps. In practice the task vectors of different tasks interfere, and the standard remedies--averaging, magnitude pruning, sign election--operate entry by entr

Read the original at arxiv.org →Open in live feed Read that day’s brief

Joint Null-Space Projection for Data-Free Model Merging

Earlier in this thread 4 items

A Margin Perspective on LoRA: Robustness to Catastrophic Forgetting and Adapter Merging (MaLoRA)

Joint Quantization and Dynamic Partitioning for Context-Aware Edge Vision Transformers

JHiMA-WebRL: A Joint Hierarchical Multi-Agent Reinforcement Learning Framework for Long-Horizon Web Tasks

When to Assist and What to Suggest: Joint Gating and Memory-Grounded Hint Generation in One Compact Decoder