Allen Institute for AI

Tin tức, mô hình và bản phát hành AI mới nhất từ Allen Institute for AI. ['MolmoAct2', 'OLMo 2', 'OLMo 2 7B', 'OLMo 3', 'OLMo 3 7B', 'Tülu 3']

Trang chính thức ↗

Tác tử 🇺🇸

What Building the Shippy Agent Taught Us: How to Build Reliable AI Agents for Operational Tasks

The Skylight (Ai2) team shared their experience developing Shippy, an AI agent for maritime awareness. The main challenge is ensuring reliability under high-stakes conditions. Shippy's architecture includes a "soul" (system prompt), skills (markdown files), and configuration. For determinism, the agent uses a CLI instead of direct API calls. Each user works in an isolated session via the Mothership platform. Agent evaluation is conducted on live data using custom scenarios; Shippy is open for testing by early users.

Allen Institute for AI

Anthropic

Hugging Face blog27.07 · 02:03

Mã nguồn mở 🇺🇸

Gigatoken: BPE Tokenizer in Rust Encodes Text at 24.53 GB/s, Up to 989x Faster Than HuggingFace Tokenizers

Stanford PhD student Marcel Roed released Gigatoken, a BPE tokenizer in Rust with Python bindings, achieving speeds of up to 24.53 GB/s on a 144-core AMD EPYC, 989x faster than HuggingFace Tokenizers and 681x faster than tiktoken. The speedup comes from a handwritten SWAR pre-tokenizer and pre-token caching.

OpenAI

Hugging Face

Google/DeepMind