【今の生成AIの勢力図がわかる図解】

精度がGPT-4にも匹敵するOSSのLlama 3が登場してから、OSSのLLMに再度スポットライトが

更に、小型化されたPhi-3も登場

GPUでなく、CPUで動く軽量・高精度のLLMが今後もっと進化して、スマホ・ウェアラブルで高性能AIが使える時代ももうすぐ

激動のAI業界

ローカルLLMの進化が本当に期待しかない

Visionまでローカルで高速で動く。
（動画は３倍なので、3分の１程度の性能だと思ってみてください。それでも十分実用性あります）

以下、評判の良かった２つのモデルをベースにしたVision ADAPTER付きモデル。

・LLaVA++ based on Phi-3
・LLaVA++ based on…

ラズパイ5でもLLM（Llama 3）あっさりと動きました

AKP'nin ulusal eğitimi özelleştirerek dincileştirme politikası.. Mutlaka engellenmeli.. öyle değil mi??!!

LLMの学習で「次の単語予測」でなく「次の複数単語予測」で学習すると,性能がかなり上がったというMetaの研究
'Better & Faster Large Language Models via Multi-token Prediction'
arxiv.org/abs/2404.19737
「サイズが大きいモデルほど有用」らしく，昔からあるアイディアが大規模学習で花開いた感

Excited to share our work at Google DeepMind!

We propose Naturalized Execution Tuning (NExT), a self-training method that drastically improves the LLM's ability to reason about code execution, by learning to inspect execution traces and generate chain-of-thought rationales 🧵👇

音は止んで

Now Qwen1.5-110B is on the top among pretrained models in Open LLM Leaderboard😉

1/ Introducing Alpha LLM: a suite of three tailor-made, unbiased #AI models designed specifically for #Web3 .

Meet Versa, Nexus, and Eclipse. Each model is crafted to meet distinct needs within the crypto ecosystem and all available via our new API.

Learn more 🧵⤵️

ローカルLLM界隈、普通に少年漫画みたいで見てるだけで楽しいのでオススメです

Released a free tool on ChatDB: Parquet AI

Query parquet files with natural language in the browser.

◆ Powered by DuckDB in the browser
◆ LLM from Groq Inc and Llama-3-70B

Here's me querying the capybara-dpo dataset from Hugging Face

Introducing Snowflake Arctic. An efficiently intelligent and truly open LLM built by Snowflake.

Another triumph for Self-Play! Self-Play Preference Optimization (SPPO) has surpassed (iterative) DPO, IPO, Self-Rewarding LMs, and others on AlpacaEval, MT-Bench, and the Open LLM Leaderboard.

Remarkably, Mistral-7B-instruct-v0.2 fine-tuned by SPPO achieves superior

本日の  #LayerX機械学習勉強会  では、LayoutLLMについて紹介しました。
その他、FastAPI Tips、MicrosoftのMLOpsホワイトペーパー、Active Learningについての論文、LLMエージェントのデザインパターンについての記事などが紹介されました。
jobs.layerx.co.jp/bdc4697b9e044d…

自分でも考えるのだけどビカーッ！バキーッ！楽しーっ！かもしれない

Solving FSD is way easier than having wipers understand sunny isn’t time to wipe. You can’t wipe sun Tesla AI! Input that into you LLM or whatever the F.

行く人います？
#sumika

No one is talking about this major LLM from China.

2 days ago, SenseTime launched SenseNova 5.0, which according to the report (translated from Chinese):

- Beats GPT-4T on nearly all benchmarks
- Has a 200k context window
- Is trained on more than 10TB tokens
- Has major

Whilst I was visiting Bloemfontein, I got a chance to attend Simphiwe Dlamini's graduation at UFS. She was my first masters student.

I supervised her LLM thesis titled: 'Authorship in Copyright Law: A Critique in the Context of the 4th IR'

So proud of you Dlamini! 💯👏🏽🙂