AgentBaton: Efficient Model Collaboration in Agentic Tasks
AgentBaton explores one-time handoffs from capable to efficient models, reporting major API cost reductions while preserving accuracy on long-horizon agent tasks.
ls ./papers | grep "#LLM evaluation"
Trending AI research papers from Hugging Face, summarized in plain language: what each paper does, why it matters and where to read more.
5 papers · #LLM evaluation
AgentBaton explores one-time handoffs from capable to efficient models, reporting major API cost reductions while preserving accuracy on long-horizon agent tasks.
EvoMemState treats personal-agent memory as evolving state, linking facts through updates and conflicts while learning how each memory should influence a request.
BestRand keeps random benchmark compression’s low cost while using probe models to reduce rare, high-error subsets across text, multimodal, agent, and VLA evaluations.
R²Orch combines self-play, adaptive task selection, and evidence-based skill withdrawal to train agent orchestrators that retain performance when skills are removed.
Co-Design Bench tests whether coding agents can turn real users’ multimodal requests into complete, working application revisions—not merely plausible partial edits.