neptune.ai

A 2-minute demo showcasing how neptune.ai supports teams that train foundation models. Haven't heard about Neptune before? TL;DR: It's an experiment tracker built to support teams that train large-scale models. Neptune allows you to: → Monitor and visualize months-long model training with multiple steps and branches. → Track massive amounts of data, but filter and search through it quickly. → Visualize and compare thousands of metrics in seconds. You get to the next big AI breakthrough faster, optimizing GPU usage on the way. If you want to learn more, visit: https://buff.ly/4cXZGep Or play with a live example project here: https://buff.ly/3WlPVQg

Calling all Kagglers: Want to increase your chances in ML competitions? Neptune's advanced experiment comparison options and lightning-fast UI are the secret weapons of Kaggle Grandmasters. Track massive amounts of experiments and iterate quickly – all for free. Check out our free program: https://buff.ly/47dzgTU #generativeai #genai #llm #ml #researchers

What AI technologies are people excited about? Prashanth Jayachandran believes quantum computing could be a game-changer for protecting high-security systems like government databases. — (Link to the full interview in the comments) #generativeai #genai #llm

When your model training runs for weeks, you need a convenient way to monitor it and share progress with stakeholders. Our Reports feature gives you this option. You can create persistent, customizable dashboards that update automatically as training progresses. Filter only the information you want to track and visualize only the metrics you are interested in. See an example report here: https://buff.ly/3YDCwVW #generativeai #genai #llm

What does it take to make AI research accessible to all? Shreyash Arya (Researcher at Max Planck Institute for Informatics) put it to the test by explaining his paper “B-cosification: Transforming Deep Neural Networks to be Inherently Interpretable" to three different audiences. See each approach in the full video: https://buff.ly/48N6xWT — Would you like to share your research with our community? Apply here: https://buff.ly/4fL6ygq

What’s ahead for AI in the next five years? Here’s Tim Pietrusky’s perspective: AI will be more than just a "nice-to-have" tool. It’s a technology that will fundamentally redefine our work. In five years (if not sooner), the impact of these changes will be clearly visible. — (link to the full interview in the comments) #generativeai #genai #llm

[New on our blog] How to Run LLMs Locally by Gabriel Gon?alves TL;DR → While many applications rely on LLM APIs, local deployment of LLMs is appealing due to potential cost savings and reduced latency. Privacy requirements or a lack of internet connectivity might even make it the only option. → The major obstacle to deploying LLMs on premises is the memory requirements of LLMs, which can be reduced through optimization techniques like quantization and flash attention. If inference latency is not a concern, running LLMs on CPUs can be an attractive low-cost option. → Libraries and frameworks like Llama.cpp, Ollama, and Unsloth help set up and manage LLMs. Best practices for building local LLM applications include abstracting the model and using orchestration frameworks or routers. — (link to the full article in the comments) #generativeai #genai #llm

When training large models - like foundation models - your training jobs run for weeks or months. You generate a lot of data that needs to be tracked and analyzed. We’re talking billions of data points. And all this data is important to track. You can’t skip some of it cause you might need to look into it at some point. Neptune ingests 100k+ data points per second. Making it possible to log every single piece of your training metadata — regardless of the scale. For now, it’s supported in Neptune Scale, our upcoming product release. You can test it yourself: https://buff.ly/4eCFUpz Soon, it will be available for everyone. #generativeai #genai #llm

Jonas NGNAWE (PhD student at Institut intelligence et données (IID) de l'Université Laval and Mila - Quebec Artificial Intelligence Institute) broke down his research paper "Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers" into three complexity levels: ? For a young learner – sharing easy basics anyone can follow. ? For a student – with more detail for those with some foundational knowledge. ? For a fellow AI researcher – going in-depth and very technical. Watch the full video for the overview at each level: https://buff.ly/3O4ZVti — Would you like to share your research with our community? You can still do it, apply here: https://buff.ly/3CkBUfo #generativeai #genai #llm

To all Polish engineers out there! We’re excited to take part and sponsor this year’s Polish Academic Championship in Collaborative Programming! You can join us online: www.live.amppz.edu.pl We start on November 17th at 8:45 am CET — We’re expanding our engineering team to build an industry-standard system for AI researchers. If you're interested in tackling complex technical challenges, come and say hi to Piotr Niedzwiedz, Adam Nie?urawski, Marcin Kierski, or Marta Hoppe ????. You can track the currently open roles here: https://buff.ly/48TGQUp