Paris booksellers refuse to move for Olympics. 💪💪

Another new SOTA Japanese LLM has just been released, weblab-10b from the University of Tokyo Matsuo Lab's Kojima Takeshi 🥳🥳! My GPU cluster is down for upgrades this weekend so adding it to Rakuda will have to wait, but on the JGLUE benchmark weblab-10b leads the pack!

Sam Passaglia We use your Rakuda Benchmark in our blog (written in Japanese.) Thank you!

Line used my benchmark to evaluate their new model 🤯🤯🤯🤯 ! Really amazing, thank you so much.

A new article for Nature by Tokyotronic discusses the race to develop home-grown LLMs in Japan! 🐪 There's even a couple quotes from me discussing Rakuda. nature.com/articles/d4158…

Biden's AI Executive Order is out and it’s terrible for US innovation. Here are some of the new obligations, which only large incumbents will be able to comply with 👇

So big announcement: thanks to the generous support from Hugging Face I am releasing the early modern ChatGPT, MonadGPT huggingface.co/spaces/Pclangl… Any question in English or French will be answered from the perspective of someone living between 1500 and 1750.

If you're in Kobe attending NLP 2024, come visit me at ELYZA's booth all week! We have cute stickers.

【お知らせ】700億パラメータの日本語LLMを開発し、グローバルモデルに匹敵する性能を達成しました。本モデルを含むモデル群を「ELYZA LLM for JP」シリーズとして順次サービス提供を開始します。 まずはデモサイトで性能をお試しください。 elyza.ai/lp/elyza-llm-f…

Today we announced our plans to deepen our investments in Japan, spanning cloud and AI infrastructure, skilling, research, and cybersecurity, as we continue partnering to accelerate the country's AI transformation.

My team at Elyza has been selected by the Japanese government to receive a major supercomputer grant to develop foundation models! Excited to put these GPUs to work :)

7th August is our 3rd TAI AAI (Tokyo AI (TAI) Advanced AI) session, focused on NLP. Come listen to researchers and engineers from the top Japanese labs presenting their work on LLMs. Apply to attend: lu.ma/iiu09leb Organizers: Kai Arulkumaran and Sam Passaglia

I had such a blast at @tokyoaijp's NLP session last night hearing from our 4 amazing speakers, @[email protected] Kakeru Hattori Ayana Niwa Mengsay Loem, and speaking to the ~100 attendees! Now I'm pumped to organize more events with Kai Arulkumaran and Ilya Kulyatin in the future :)

The Flood risk map for Tokyo is art 🌊

New Differential Transformer paper (Tianzhu Ye ++) is really cool: they make attention heads differential, computing two attention maps per input and subtracting them. This improves performance by cancelling out noise, like a humbucking guitar. arxiv.org/abs/2410.05258

LLM-jp Chatbot Arena を公開しました。 chatbot-arena.apps.llmc.nii.ac.jp LLM-jp-3 172Bを含む計10モデルと会話できます。収集したデータは LLM-jp から公開予定です。 2/11 (火) 9:00 までの稼働を予定しています。短い期間ですが、ぜひお試しください。

日英4.4T tokensで学習した日本語ModernBERTを公開しました！！ 系列長8192、語彙数は日英10万、パラメータ数130Mながら既存largeモデルと同等以上の性能があります 12データセットによる既存BERT系モデルの網羅的な評価も行いましたので、そちらもぜひ！！

Had so much fun hosting this panel “Is Scale Enough?” on Algorithms Day for the Open Problems for AI Summit in Tokyo 🇯🇵 with Kai Arulkumaran Emtiyaz Khan Jad Tarifi Sam Passaglia Masanori Koyama 🤖 we discussed definitions of AGI, and how we can advance algorithmic research 🚀