Engineering
FLEET: From Logits Entropy to Enhanced Trajectories in Text Generation
Introduces FLEET, a decoding-time method that adds a memory mechanism to LLM sampling to counter the diminishing returns of memoryless…
Engineering
Six Layers Less: Encoder Pruning for Whisper with Label-Free Recovery
Presents a method for pruning Whisper’s ASR encoder (rather than the more commonly pruned decoder) by ranking layers via leave-one-layer-out…
Engineering
EmbodiedSWE: Coding Agents for Long Horizon Dexterous Robotics
Studies whether coding agents can generate scalable supervision for training general robot policies by introducing EmbodiedSWE-Bench, a simulation benchmark covering…
Engineering
EvoOntology: A Self-Evolving Ontology Layer for Data Agents
EvoOntology is a self-evolving ontology layer that lets data agents actively query and update a schema/content/tool-layer ontology exposed as an…
Engineering
CodeMidas: Turning Existing Codebases into Scalable RL Environments for Coding Agents
CodeMidas is an agentic pipeline that converts the implemented functionality inside open-source codebases into executable reinforcement-learning environments, using only source…
Engineering
Code2Skill: Synthesizing Verified Agent Skills at Scale from Source Code
Code2Skill is a fully automated pipeline that mines source code repositories to produce implementation-anchored agent skills, verifying each one through…
Engineering
RecreationWorld: Scalable, Verifiable Environments for Hybrid GUI-and-Code Computer-Use Agents
RecreationWorld is a five-platform (Ubuntu, macOS, Windows, Android, Web) framework that trains hybrid computer-use agents to both explore interfaces and…
Engineering
IntBMoE: A Block-Conditioned Mixture-of-Experts Architecture Deployed in Production Recommendation Serving
IntBMoE is a Mixture-of-Experts design that decouples per-token participation, compute cost, and memory cost by pairing dense expert composition with…
Engineering
OmniVChat: A Data Engine, Benchmark, and RL Training Recipe for Native Audio-Visual Dialogue Models
OmniVChat defines and studies native audio-visual dialogue, where an omni model receives simultaneous audio and video (with the query embedded…
Engineering
Designer-RSI: A Self-Improving Procedural Memory for Agentic Graphic Design
Designer-RSI is a continual-adaptation framework in which a frozen frontier model drives professional design software through 230+ tools while an…
Engineering
OmniVBench: A Factor-Grounded Benchmark and Dataset for Omni Reference-to-Video Generation
OmniVBench addresses the lack of adequate evaluation for omni reference-to-video generation, where models must preserve, disentangle, and correctly route many…
Engineering
BI-Agent and BI-Bench: A Tool-Augmented Agent and Benchmark for End-to-End Business Intelligence
The authors introduce BI-Bench, the first benchmark systematically testing whether LLMs can answer business-intelligence questions end-to-end (table discovery, transformation, joins,…