Allen Institute for AI

Latest AI news, models and releases from Allen Institute for AI. ['MolmoAct2', 'OLMo 2', 'OLMo 2 7B', 'OLMo 3', 'OLMo 3 7B', 'Tülu 3']

Official site ↗

Open Source 🇺🇸

LeRobot v0.6.0: World Model-Based Policies, New VLA, Benchmarks and CLI for Evaluation

LeRobot v0.6.0 has been released with support for three world model-based policies (VLA-JEPA, FastWAM, LingBot-VA), new VLA (GR00T N1.7, MolmoAct2, EO-1, EVO1, Multitask DiT) and a unified API for reward models (Robometer, TOPReward). Six new simulation benchmarks have been added and integrated into lerobot-eval, a CLI lerobot-rollout with DAgger corrections, depth support, automatic language annotation, and up to 2x data loading speedup.

Hugging Face

NVIDIA

Allen Institute for AI

Alibaba/Qwen

Hugging Face blog27.07 · 04:06

Agents 🇺🇸

What Building Shippy Taught Us About Building Reliable AI Agents for Operational Tasks

The Skylight team at Ai2 built Shippy, an AI agent for real-time maritime domain awareness. The key challenges were ensuring reliability, correct tool use, and user data isolation. Shippy uses a deterministic CLI tool, sandboxed sessions via Mothership, and a custom evaluation system with subject-matter expert rubrics.

Allen Institute for AI

Anthropic

Hugging Face blog27.07 · 02:03

Show more ↓