Panel’s Post

Learn about new tools you can use for human-post tracking by following along Andreas Naoum's latest tutorial, which walks us through a complete workflow based on MediaPipe and Rerun.

Short intro by Rubens on how to build a knowledge graph with some embeddings and cool graph visualizations. #knowledgegraph #graphdatabase #vertexai #neo4j

How to efficiently 🚀 run BERT model from HuggingFace? How do various BERT variants, such as DistilBERT and RoBERTa, stack up in terms of inference throughput 🥊? ⭐ Check out my latest blog post https://lnkd.in/e4uMnvR6

🚀 Here is the third post in my "I Learn, You Learn" series! This time, I delve into the Generation aspect of Retrieval-Augmented Generation (RAG), which is essential for creating accurate and informative AI responses.   📚 Basics Of RAG (Retrieval Augmented Generation) — GENERATION   In this article, I explore how to effectively incorporate retrieved documents into the context of large language models (LLMs) to enhance their response generation capabilities:   1. Adding Docs to Context Window a. Context Window: The segment of text that an AI model considers when generating a response. For LLMs, this window is limited in size, so only the most relevant information is included. b.Incorporation: Retrieved documents are added to the context window alongside the original query or prompt, enriching the context to help the model generate more accurate and informative responses. Think of it like providing an AI with a relevant chapter from a book along with a question, so it has more detailed information to give a better answer. 2. Connecting Retrieval with LLMs via Prompt a. Retrieval: The system first retrieves the most relevant documents based on the similarity search. b. Prompt Construction: These retrieved documents are combined with the user’s query to form a comprehensive prompt fed into the LLM. c. Response Generation: The LLM uses this detailed prompt, which includes both the query and the additional context from the retrieved documents, to generate a more informed and accurate response. Imagine you’re asking a librarian a question, and they provide you with a specific book or article along with their answer to give you a more complete and accurate response.   Check out the full article here: https://lnkd.in/eHJXwFhT Stay tuned for more articles in the "I Learn, You Learn" series. Let's continue this learning journey together!   #AI #MachineLearning #RAG #ArtificialIntelligence #Generation #DataScience #TechInnovation #MediumArticle #ILearnYouLearn #ContextWindow

Build a playlist recommendation system using an end-to-end candidate retrieval workflow with #VertexAI → https://goo.gle/3GLMoDo 💡 Learn all about: 🟠 The evolution of retrieval modeling 🟠 Developing custom two-tower encoders 🟠 Framing a playlist-continuation use-case ... and more!

AI based #recommendersystems enhance several use cases and applications across business scenarios. Vertex AI platform makes it easier to build, deploy and manage at scale. Thanks for sharing Aaron Hanlon! #googlecloud #vertexai

RoPE scaling is now available in 🤗 text-generation-inference 🚀 What does this mean? You can now efficiently serve Llama 2 or GPT-NeoX models with much larger sequence lengths (think 32k or more!), with and without fine-tuning 🔥 You can get it from `main`, or wait for the next TGI release. Models fine-tuned with RoPE scaling are loaded correctly without any additional input 💪 If you'd like to enable dynamic scaling instead, check the PR linked above or the CLI's --help. In other words, models with RoPE scaling like Enrico Shippole's LLongMA 2 (8k or 16k tokens) and Together AI 32k Llama 2 can now be loaded in TGI 💛 I've added the link to these models in the comments, as well as further RoPE scaling references. All the credit goes to Nicolas Patry, who brought it to life! Good luck with your GPU RAM 🫡 https://lnkd.in/dbtZNpi4

Are you a #PowerPages fan? Check out our latest #PowerTips to learn about the new features in this release wave! #PowerPlatform

A great method to increase the context length limit of your LLMs. All it takes is one parameter in your CLI or one parameter in your Transformer’s from_pretrained function!

Create a presentation for your exam class with a few sentences and a link to the spec? One of the workload saving features of AI, find out about this and more in our AI in Education course, pre-register here: https://buff.ly/47GZ75I video from the course: