LatentSkill converts textual skills to weight-space adapters for LLM agents
LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents
Hugging FaceTrending 13
Summary
LatentSkill converts textual skills into plug-and-play LoRA adapters through a pretrained hypernetwork, improving ALFWorld success by 21.4 points on seen splits while using 64.1% fewer prefill tokens compared to in-context skill baselines.