WeAct: Dynamic Weight Switching as Agent Actions
We introduce WeAct (Weight-level Internal Actions), which trains independent capability-specific LoRA adapters over a frozen backbone and adds adapter selection to the agent's action space.
ProofPaper ↗
Key points
- Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories.
- Updating shared parameters to learn new capabilities can degrade previously acquired ones, resulting in catastrophic forgetting.
- WeAct also supports modular continual adaptation, by adding a new code computation adapter to expand the agent's capabilities and updating the existing writing adapter to improve report quality.
- Both updates leave the backbone and other adapters unchanged, supporting modular continual adaptation in agents.
Sources (1)
- [1]WeAct: Dynamic Weight Switching as Agent ActionsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 05:59 PM
We introduce WeAct (Weight-level Internal Actions), which trains independent capability-specific LoRA adapters over a frozen backbone and adds adapter selection to the agent's action space.
Long-horizon agents must learn heterogeneous capabilities from interleaved trajectories.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026unslothai/unsloth v0.1.909-beta: DeepSeek-V4.1 & FLUX-3
- Oct 19, 2026The Model Swap Workshop: GPT-5.5 vs Kimi vs DeepSeek vs Mistral
- Oct 19, 2026TaQFormer: A Treatment-as-Query Transformer with Data-Fusion for Multidimensional Treatment Modeling in Coupon Allocation
- Oct 19, 2026Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
- Oct 19, 2026Scaling Parameter-Efficient Fine-Tuning for Brain Encoding
- Oct 17, 2026Peer Pressure: Execution-Verified Co-Evolution of Attacker and Defender Populations for Code Security