PulseAugur
EN
LIVE 02:06:53
한국어(KO) Genex는 MIT 라이선스의 오픈소스 AI 게임 개발 데스크톱 앱이다. 여러 AI 에이전트와 게임 화면을 이해하는 시각 평가자를 활용해 브라우저용 Three.js 게임을 만들고, 빌드 결과를 바탕으로 도구와 스킬을 개선한다. Claude·ChatGPT 구독이나 로컬 모델을 사용할 수

AI2 releases Olmo-core 3 for faster MoE model training · 2 sources tracked

By PulseAugur Editorial · [2 sources] ·

Allen Institute for Artificial Intelligence (AI2) has released Olmo-core 3, an open infrastructure designed for large-scale Mixture of Experts (MoE) model training. This new system enhances throughput by keeping experts resident on GPUs, achieving over 52,000 tokens per GPU per second for a 47B parameter model, a 2.7x improvement over previous implementations. Olmo-core 3 also demonstrated a 21% increase in training throughput with MXFP8, while reducing memory usage. This infrastructure is intended as the foundation for future Olmo models and is available for researchers. AI

IMPACT Enhances efficiency and accessibility for training large AI models, potentially accelerating research and development in MoE architectures.

RANK_REASON Release of new open-source infrastructure for large-scale AI model training by a prominent research institute.

Read on Mastodon — fosstodon.org →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

AI2 releases Olmo-core 3 for faster MoE model training · 2 sources tracked

How we ranked this

Signal score
11 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Significant
Release of new open-source infrastructure for large-scale AI model training by a prominent research institute.
Source corroboration
2 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
Topics
infra, model release
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [2]

  1. Mastodon — fosstodon.org TIER_1 한국어(KO) · [email protected] ·

    Ai2 has released 'Olmo-core 3', an open infrastructure for large-scale MoE training. By keeping experts resident on GPUs, it has increased throughput compared to existing implementations, processing 52,000 tokens per second per GPU on a 47B model, approximately 2.7 times faster than before. A 1.2 trillion parameter model on 512 NVIDIA B300

    Ai2가 대규모 MoE 학습을 위한 오픈 인프라 ‘Olmo-core 3’를 공개했다. 전문가를 GPU에 상주시켜 기존 구현보다 처리량을 높였으며, 47B 모델에서 GPU당 초당 5만2천 토큰을 처리해 이전보다 약 2.7배 빨라졌다. 1.2조 파라미터 모델을 512개 NVIDIA B300 GPU에서 시험했고, MXFP8 적용 시 학습 처리량은 21% 증가, 메모리는 103GiB에서 95GiB로 감소했다. 차세대 Olmo의 기반이며 연구자도 자유롭게 활용할 수 있다. https:// allenai.o…

    LINKS allenai.org/…/olmocore3
  2. Mastodon — fosstodon.org TIER_1 한국어(KO) · [email protected] ·

    Genex is an open-source AI game development desktop app with an MIT license. It creates Three.js games for browsers using multiple AI agents and a visual evaluator that understands game screens, and improves tools and skills based on build results. Claude·ChatGPT subscriptions or local models can be used.

    Genex는 MIT 라이선스의 오픈소스 AI 게임 개발 데스크톱 앱이다. 여러 AI 에이전트와 게임 화면을 이해하는 시각 평가자를 활용해 브라우저용 Three.js 게임을 만들고, 빌드 결과를 바탕으로 도구와 스킬을 개선한다. Claude·ChatGPT 구독이나 로컬 모델을 사용할 수 있으며, 현재 macOS·Linux에서 이용 가능하다. Unity·Unreal Engine 지원은 준비 중이다. https:// genex.games/desktop # ai # gamedev # opensource …

    LINKS genex.games/desktop

RELATED ENTITIES

RELATED TOPICS