Ethan B. Holland

Over 56,100 manually organized AI links and counting

a sunny day with blue skies. a visitor map of an apple orchard named "Open Source" lays on a wooden table. --chaos 40 --ar 4:3 --style raw --personalize 9zxyhz8

Open Source AI News: Week Ending 06/14/2024

a sunny day with blue skies. a visitor map of an apple orchard named “Open Source” lays on a wooden table. –chaos 40 –ar 4:3 –style raw –personalize 9zxyhz8

Cohere

Cohere CEO Aidan Gomez sees AI’s pathway to profitability – The Verge

https://www.theverge.com/24173858/ai-cohere-aidan-gomez-money-revenue-llm-transformers-enterprise-stochastic-parrot

Databricks

Databricks + Tabular | Databricks Blog

https://www.databricks.com/blog/databricks-tabular

Introducing Shutterstock ImageAI, Powered by Databricks: An Image Generation Model Built for the Enterprise – Databricks

https://www.databricks.co

Hugging Face

“I’m thrilled to share that the Argilla team is now part of the @huggingface family! 🔥 It’s been a year of incredible organic collaboration, and I’m so proud of welcoming them and everything accomplished. Datasets go brrr 

Meta/Llama

“Was toying around with LLM model eval that run well on a laptop. Turns out Llama 3 8B Instruct is a pretty good evaluator that runs on a MacBook Air. I got a pretty high 0.8 correlation with GPT-4 scores. Standalone notebook here if you want to give it a try: 

SEO is dead: AI will rescan the internet and make duplicate content worthless. – Ethan B. Holland – https://ethanbholland.com/2024/07/05/seo-is-dead-ai-will-rescan-the-internet-and-make-duplicate-content-worthless/

“What If We Recaption Billions of Web Images with LLaMA-3? Web-crawled image-text pairs are inherently noisy. Prior studies demonstrate that semantically aligning and enriching textual descriptions of these pairs can significantly enhance model training across various 

“What If We Recaption Billions of Web Images with LLaMA-3 ? – Finetunes a LLaVA-1.5 and recaptions ~1.3B images from the DataComp-1B dataset – Opensources the resulting dataset data: 

“”What If We Recaption Billions of Web Images with LLaMA-3?”🤯 And the results confirm that this enhanced dataset, Recap-DataComp-1B generated this way, offers substantial benefits in training advanced vision-language models. For discriminative models like CLIP, we observe 

[2406.08478] What If We Recaption Billions of Web Images with LLaMA-3?

https://arxiv.org/abs/2406.08478

Mistral

Paris-based AI startup Mistral AI raises $640M | TechCrunch

Paris-based AI startup Mistral AI raises $640M

NVIDIA

Nemotron-4 340B | Research

We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base, Nemotron-4-340B-Instruct, and Nemotron-4-340B-Reward. Our models are open access under the NVIDIA Open Model License Agreement, a permissive model license that allows the distribution, modification, and use of the models and their outputs. 

https://research.nvidia.com/publication/2024-06_nemotron-4-340b

NVIDIA Releases Open Synthetic Data Generation Pipeline for Training Large Language Models | NVIDIA Blog

https://blogs.nvidia.com/blog/nemotron-4-synthetic-data-generation-llm-training

“Nemotron-4-340B is released today! * Base, Instruct, Reward models * Permissive license * Great for Synthetic Data Generation * Designed to help others build their own models * Sized for inference on 8 NVIDIA H100 GPUs * Competitive across many tasks” / X

“Nemotron -4-340B-Reward: * Best reward model currently available * 

“Nemotron-4-340B-Base: * Trained for 9T tokens on 6144 H100 GPUs * Using Megatron-Core at 41% MFU * 96 layers, 18432 hidden state * GQA, Squared ReLU 

“Nvidia presents HelpSteer2 Open-source dataset for training top-performing reward models High-quality preference datasets are essential for training reward models that can effectively guide large language models (LLMs) in generating high-quality responses aligned with human 

Phi

“Introducing Phi-3 WebGPU, a private and powerful AI chatbot that runs locally in your browser, powered by 🤗 Transformers.js and onnxruntime-web! 🔒 On-device inference: no data sent to a server ⚡️ WebGPU-accelerated (> 20 t/s) 📥 Model downloaded once and cached Try it out! 👇 

Qwen

“Alibaba’s open-source Qwen 2-72B model moved into the top spot on Hugging Face Open LLM Leaderboard This ranks the model ahead of Mixtral and Llama-3 across a range of benchmarks Pretty wild to continue to see the pace of open-source AI 

Other Open Source News

“HELM MMLU v1.4.0 is out! Lots of new models added: Yi Large, OLMo 1.7, Command R(+), Gemini 1.5 {Flash,Pro}, Mistral Instruct v0.3, GPT-4 Turbo 2024-04-09, Qwen {1.5, 2}. Usual suspects at the top; notably, the top open-weight model is now Qwen 2 Instruct, surpassing Llama 3. 

“Not Llama 3 405B, but Nemotron 4 340B! @nvidia just released 340B dense LLM matching the original @OpenAI GPT-4 performance for chat applications and synthetic data generation. 🤯 NVIDIA does not claim ownership of any outputs generated. 💚 TL;DR: 🧮 340B Paramters with 4k 

“Mixture-of-Agents Boosts Open-Source LLM Capabilities 🤔Can combining the diverse expertise of open and closed-source LLMs enhance task capabilities? Mixture-of-Agents (MoA) proves we can! @togethercompute’s study shows that an MoA setup, using exclusively open-source LLMs, 

“Multimodal Table Understanding Introduces Table-LLaVa 7B, a multimodal LLM for multimodal table understanding. Competitive with GPT-4V and significantly outperforms existing MLLMs on multiple benchmarks. They also develop a large-scale dataset MMTab, covering table images, 

“Stability AI released the open model weights for Stable Diffusion 3 Medium. The 2B parameter text-to-image model offers “advanced photorealism, prompt understanding, and typography capabilities” 

Stable Diffusion 3 Medium — Stability AI

https://stability.ai/news/stable-diffusion-3-medium

“And SD3 by @StabilityAI is already the #1 trending model on HF! 

“The US is going to lose its leadership in AI if it doesn’t support more open research and open-source AI!” / X

“Is that what we call Bingo? 🎯 “Samba = Mamba + MLP + Sliding Window Attention + MLP stacking at the layer level.” => infinite context length with linear complexity Samba-3.8B-instruct outperforms Phi-3-mini across all benchmarks using the same dataset (trained on 3.2 

Together MoA — collective intelligence of open-source models pushing the frontier of LLM capabilities

https://www.together.ai/blog/together-moa

“Today, we’re thrilled to announce the open weights for Stable Diffusion 3 Medium, the latest and most advanced text-to-image AI model in our Stable Diffusion 3 series! This new release represents a major milestone in the evolution of generative AI and continues our commitment to 

“Prompting in prod will be dead in a few years. You get better performance, more reliable control, and cheaper inference by using a fine-tuned task-specific adapter. You can deploy this stack today using @OpenPipeAI, which makes fine-tuning just as easy as prompting.” / X

“Dropping a new open dataset, Character Codex! This contains data on 15,939 characters from a wide variety sources, from anime to historical figures, scientists to pop icons both fictional and non-fictional! Download on HuggingFace: 

“We are excited to announce that @argilla_io joins the @huggingface Family! 🤗 Over the past year and a half, we’ve collaborated on including releasing the largest synthetic open RLHF dataset to date. 👀 Synthetic data is on the rise and will be the key to enabling companies to 

“Excited to announce @LaminiAI Memory Tuning, a new research breakthrough! 🎉 ◽95%+ accuracy, cutting hallucinations by 10x ◽Turns any open LLM into a 1M-way adapter MoE (paper & Lamini-1 model weights on @Huggingface) ◽Fortune 500 customer case study on how they memory-tuned” / X

“Introduce HumanPlus – Shadowing part Humanoids are born for using human data. We build a real-time shadowing system using a single RGB camera and a whole-body policy for cloning human motion. Examples: – boxing🥊 – playing the piano🎹/ping pong – tossing – typing Open-sourced! 

“Cognitive Computations presents: Dolphin-2.9.3-qwen2-0.5b and Dolphin-2.9.3-qwen2-1.5b Two tiny Dolphins that still pack a punch! Run it on your wristwatch or your raspberry pi! We removed the coding, function calling, and multilingual, to let it focus on instruct and 

“8B-parameter Mamba-2-Hybrid exceeds the 8B-parameter Transformer on all 12 standard tasks we evaluated (+2.65 points on average) and is predicted to be up to 8× faster when generating tokens at inference time. 🤯 📌 The hybrid model also demonstrates strong long-context 

“Web Search, URL Fetcher, Document Parser, Imagine Generation and Editing, and Calculator: HuggingChat now integrates six tools. Handy to pi** off your Italian friends (or any more useful example you can think of) 

Heads up! You’ve scrolled to the end of this category. There may have been just one or two links (above), so go back up and double check to be sure you didn’t quickly scroll down past it.

Be Sure To Read This Week’s Main Post:

This week’s executive overview and top links are here:

AI News #37: Week Ending 06/14/2024 with Executive Summary and Top 7 Must-Read Links

The post you just read is an deep dive extension of my weekly newsletter, This Week In AI, an executive summary of the top things to know in AI. Each week, I create an accessible overview for laypeople to feel confident they are conversant with the week’s AI developments. I include a curated list of must-click links of the week, to offer everyone a hands-on opportunity to explore the most intriguing updates in artificial intelligence across various categories, including robotics, imagery, video, AR/VR, science, ethics, and more. Beyond the overview, I post these topic-based deeper dives (below). If you haven’t read this week’s overview, I recommend starting there.

Credits/Sources

Most of these weekly links come from just a few prolific oversharing sources. Please follow them, as they work hard to find the news each week and they make it a lot easier for me to compile.

For previous issues, please visit the archives!

Thanks for reading!

Leave a Reply

Trending

Discover more from Ethan B. Holland

Subscribe now to keep reading and get access to the full archive.

Continue reading