Darius Alnaddaf

Kraken Update 🐙 : Kraken Model now Online on Hugging Face https://lnkd.in/e2miq2Jy Have fun testing it 🤗 ⏩ Experts of Kraken ⏺ defog-llama-3-sqlcoder-8b: Leading model for SQL generation – outperforms GPT-4 Turbo. ⏺ gorilla-llm-gorilla-openfunctions-v2-7b: Leading model for function callings – outperforms GPT-4. ⏺ ise-uiuc-Magicoder-S-DS-6.7B: Top Python model in its size. ⏺ VAGOsolutions-Llama-3-SauerkrautLM-8b-Instruct: Strongest Llama-3-8b Fine-Tune for the German language. #MachineLearning #GPT #LLM #Generativeai #OpenSource

14

2 commentaires

Your Contribution for the start of the week 📆: botpress 👇 Co-founded by Sylvain Perron and Justin Watson, Botpress allows developers to build chatbots without a team of data scientists and machine learning experts. Here’s why you should contribute to this repo: 1️⃣ By contributing to botprocess, you can dive into complex bot development using the Botpress SDK/CLI and your contributions have the chance of positively impacting the experience of hundreds of thousands of developers. Since maintainers take extra care in getting back to you (with a ~100% reply rate and an average of 1 day to answer your queries), your contributions will get the attention they deserve. 2️⃣ Nearly 50% of contributors submit a second PR. This means the repo has a supportive environment for building upon past contributions. If you enjoy working on this repo, chances are you can continue to work on it! ⚡️ 3️⃣ Botpress is an accessible repo. At Quira, we rank Developers by their DevRank (more on this here: https://lnkd.in/eNtrs97F). We found that the median developer working on the repo belongs to the ~85th percentile of GitHub developers. In comparison, many of the open source companies we reported on have had a median contributor level of ~95th percentile. This means that you have a real opportunity to contribute. ⚡️ You can check more of replicate/cog's open-source stats on Quira (https://lnkd.in/efemmXwH) or access the GitHub page https://lnkd.in/eD8SdZBH 📖

8

Want to test Meta, Google, and Mistral models all on one platform ? **Groq.com** Why Groq? Because speed matters. GroqCloud supports a diverse array of models, enabling you to select the best for your needs: ➡ llama3-8b-8192 by Meta (Context Window: 8,192 tokens) ➡ llama3-70b-8192 by Meta (Context Window: 8,192 tokens) ➡ mixtral-8x7b-32768 by Mistral (Context Window: 32,768 tokens) ➡ gemma-7b-it by Google (Context Window: 8,192 tokens) Groq's commitment to GenAI inference speed is transforming real-time AI applications with its LPU. What is an LPU? ➡ Language Processing Unit, is a revolutionary processing system designed specifically for LANGUAGE applications. Unlike traditional CPUs (Central Processing Units) and GPUs (Graphics Processing Units), which handle a broad range of tasks. LPUs are optimized for the rapid processing demands of LLMs This specialized architecture allows for more efficient handling of sequential data, significantly reducing the time per word processed. ➡ That means you get faster outputs You can run Llama-2 70B at over 300 tokens per second per user!

24

Check out our upcoming digital workshop with Rockset and elevate your data visualization skills 🚀📊 🗓️: May 29th at 10am PT 👩‍💻👩‍💻: Sofia Jakovcevic & Annette Chen Tailored for data pros and developers, here's what you’ll learn: 🔗Data Ingestion: How to use Explo and Rockset to ingest data from various streaming sources. 🎯 Query with Precision: How to execute real-time queries with Rockset’s SQL capabilities. 📈 Visualize Instantly: How to create interactive customizable dashboards and data reports with Explo. Don’t miss this chance to transform streaming data into actionable insights. Register here👉 https://lnkd.in/gCw9JkdV --- #DataVisualization #Explo #Rockset #RealTimeData #Analytics

12

1 commentaire

What are the biggest challenges in Spatial Data Science 🤔? We want to hear from you 🫵 Share your experience and enter to win a $100 Amazon Voucher 💻📋: https://hubs.ly/Q02vjMbZ0

7

📢 Hugging face is directly integrated with GenAI developers favorite tool "langchain"🦜 Get set go!🚀 💡 pip install langchain-huggingface https://lnkd.in/ddHreVrs #huggingface #GenerativeAI #langchain

22

🛡Julien M. has a few words about scikit-learn for you: de-facto #standard, #gamechanger, #userfriendly, #accessible, #inspiring, ❤️ #thankyou, #democratizeAI Source: 🛡Julien M. Want to make your own scikit-learn #appraisal and help us spread the love? Fill in this form: https://lnkd.in/eB2V6hWN ::probabl. is recruiting, check our open positions: https://lnkd.in/eUJ6UWiM #OwnYourDataScience

28

🚀 *Automatic Data Curation for Self-Supervised Learning: A Clustering-Based Approach* by researchers from FAIR at Meta, INRIA, Université Paris Saclay, and Google. This innovative approach could transform how we handle large-scale data for machine learning. The Problem: Manually curating datasets for self-supervised learning is a huge challenge. Imagine trying to create a dataset for training an image recognition model. You’d need to collect millions of images, label them accurately, and ensure a diverse representation of concepts (like different breeds of dogs or types of vehicles). This process is not only incredibly time-consuming but also prone to human error and bias, which can seriously affect the model's performance. The Solution: The authors present a clustering-based method to automate the curation process. Here’s a quick rundown of how it works: - Successive and Hierarchical K-Means Clustering: This method is applied to a large and diverse data repository to create clusters that uniformly represent different data concepts. - Balanced Sampling: Data points are sampled from these clusters to ensure a balanced representation of concepts, which is crucial for avoiding biases towards more frequent concepts in the dataset. What They Found: - This automatic curation method significantly improves performance on various benchmarks, showing that it can even enhance already curated data pools. - The approach is versatile, applicable to different domains including web-based images, satellite images, and text data. - The highly curated dataset can then be used for self-supervised learning to build a model tailored for your specific domain, making it a powerful tool for industry-specific applications. Why It Matters: By leveraging this method, we can scale up the creation of high-quality datasets, paving the way for more robust and generalized models in machine learning. 🔗 See more details in paper: https://lnkd.in/eADUJy6d 🔗 Check out their code on GitHub (coming soon): https://lnkd.in/eRR769JB

15

🚀 New Blog Post ! 🚀 Discover how to build an advanced #RAG #chatbot with Neo4j and LangChain. This guide covers everything you need, from setting up Neo4j for knowledge storage to using Langchain for NLP and OpenAI's LLMs for text generation. Perfect for developers and AI enthusiasts looking to create more natural and informative chatbots. Read now: https://lnkd.in/dxgYrChd #AI #Chatbots #NLP #Neo4j #Langchain #OpenAI #FutureSmartAI

281

7 commentaires

🚀 𝗡𝗲𝘄𝘀 𝗳𝗿𝗼𝗺 𝘁𝗵𝗲 𝗚𝗲𝗻𝗔𝗜 𝗯𝗮𝘁𝘁𝗹𝗲𝘀: 𝗖𝗹𝗮𝘂𝗱𝗲 3.5 𝗦𝗼𝗻𝗻𝗲𝘁 𝗶𝘀 𝘁𝗵𝗲 𝗟𝗟𝗠 𝘁𝗼 𝗯𝗲𝗮𝘁 Claude.ai is excited to announce the launch of 𝗖𝗹𝗮𝘂𝗱𝗲 3.5 𝗦𝗼𝗻𝗻𝗲𝘁, the first in their new Claude 3.5 model family. This release sets a new benchmark for AI intelligence, outperforming both competitor models and Claude 3 Opus in numerous evaluations—all while maintaining the speed and affordability of the mid-tier Claude 3 Sonnet. 🎉 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: - 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲: Claude 3.5 Sonnet excels in graduate-level reasoning (GPQA), undergraduate-level knowledge (MMLU), and coding proficiency (HumanEval), showing marked improvements in understanding nuance, humor, and complex instructions. - 𝗖𝗼𝘀𝘁-𝗘𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲 𝗮𝗻𝗱 𝗙𝗮𝘀𝘁: Operating at twice the speed of Claude 3 Opus, Claude 3.5 Sonnet is priced at $3 per million input tokens and $15 per million output tokens, with a 200K token context window—making it ideal for complex, context-sensitive tasks. - 𝗦𝘁𝗮𝘁𝗲-𝗼𝗳-𝘁𝗵𝗲-𝗔𝗿𝘁 𝗩𝗶𝘀𝗶𝗼𝗻: This model surpasses its predecessors in visual reasoning, effectively interpreting charts, graphs, and transcribing text from imperfect images. 🌟 𝗪𝗵𝘆 𝗖𝗵𝗼𝗼𝘀𝗲 𝗖𝗹𝗮𝘂𝗱𝗲 3.5 𝗦𝗼𝗻𝗻𝗲𝘁? - 𝗨𝗻𝗺𝗮𝘁𝗰𝗵𝗲𝗱 𝗦𝗽𝗲𝗲𝗱 𝗮𝗻𝗱 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆: With performance metrics far exceeding those of competing models, Claude 3.5 Sonnet is ideal for complex, context-sensitive tasks. - 𝗘𝗻𝗵𝗮𝗻𝗰𝗲𝗱 𝗩𝗶𝘀𝗶𝗼𝗻 𝗖𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀: Superior in visual reasoning, it effectively interprets charts, graphs, and transcribes text from imperfect images. - 𝗗𝘆𝗻𝗮𝗺𝗶𝗰 𝗪𝗼𝗿𝗸𝘀𝗽𝗮𝗰𝗲 𝘄𝗶𝘁𝗵 𝗔𝗿𝘁𝗶𝗳𝗮𝗰𝘁𝘀: Users can generate, edit, and build upon AI-generated content in real-time, enhancing collaboration and productivity. Read more about it here: https://lnkd.in/dbXVHtad #AI #Innovation #ClaudeAI #ArtificialIntelligence #GenAI

10

Baal, our Bayesian Active Learning library is working on a major version and we want to know more about you! If you use Baal for Active Learning, Uncertainty Estimation or Bayesian Deep Learning, we would **love** to talk to you! 😎 In more detail, we want to understand when our users use our library and how. You can take a spot in our Calendly:

14

🤖 Is your data AI-ready? If the issues explained in this carousel sound familiar, join Laurent Dresse ☁ for our upcoming webinar to learn more about how a data catalog can help you optimize your AI data governance. Register here: https://lnkd.in/d8Dp_s4H #webinar #ai #datacatalog #datagovernance

23

If you want to understand why LLMs are not the solution for AGI (Artificial General Intelligence), I highly recommend listening to Lex Fridman podcast with Yann LeCun : Meta AI, Open Source, Limits of LLMs, AGI & the Future of AI T͟L͟D͟R͟: 1. LLMs do not understand the physical world 2. LLMs have not the ability to remember 3. LLMs have no persistent memory 4. LLMs are not able to reason and to plan

27

5 commentaires

Shower thoughts 💭: The training and performance of Llama3 demonstrate a shift in the scaling law for language models. Instead of simply increasing the model size to achieve lower loss, the focus is now on keeping the model size constant while adding more high-quality training data. Llama3, with 8 billion parameters and 6-7 times more data, matches the performance of early GPT3.5 models with over 100 billion parameters. The logic behind this is that as large models are put into practical use, the inference cost becomes more significant than the training cost. The high performance achieved by large models is ultimately overshadowed by the inference cost. In this context, the biggest pain point of federated learning in decentralization becomes less relevant, as pursuing more data is more cost-effective than pursuing larger models. Many emerging AI+Crypto companies still emphasize training larger decentralized models, which is actually an outdated concept. FLock.io on the other hand, recognized this shift even before the emergence of Llama3 and has been focusing on federated learning from the beginning. #AI #MachineLearning #FederatedLearning #Llama3 #LanguageModels

18

looks like I'm not the only one trying to mine arxiv I'm using pymupdf, but txtai has some examples scraping pdfs I like the layout detection this article has plus the ability to remove references rag over arxiv has become my new pet project (sorry stocks, chasing you was exhausting and not directly in line with the latest ml trends). eventually though, whatever I cook up for arxiv will be applied to 10ks https://lnkd.in/eJf8k8_z

1

🔬 Exciting Breakthrough in AI Interpretability by Anthropic ! 🔬 Eight months ago, our team embarked on a challenging journey to scale sparse autoencoders for recovering monosemantic features from transformers. Today, we're thrilled to share a significant milestone: we've successfully extracted high-quality features from Claude 3 Sonnet, Anthropic's medium-sized production model! Our research has unveiled a fascinating diversity of highly abstract features that both respond to and influence behaviors. Some key findings include: - Features representing famous people, countries, cities, and type signatures in code. - Multilingual and multimodal features, bridging text and images across languages. - Abstract and concrete instantiations of ideas, such as code vulnerabilities and discussions on security. 📊 Key Results: 1. Interpretable Features: Sparse autoencoders are producing clear, interpretable features even for large models. 2. Scaling Laws: We've utilized scaling laws to guide our training, enhancing efficiency and effectiveness. 3. Abstract Generalization: The features generalize across multilingual, multimodal, concrete, and abstract references. 4. Systematic Relationships: A systematic relationship between concept frequency and dictionary size needed for feature resolution. 5. Influence on Behavior: These features can be used to steer large models, influencing behavior significantly. Importantly, the findings also touch on crucial safety concerns, revealing features related to deception, sycophancy, bias, and dangerous content. This breakthrough marks a significant step forward in AI safety and interpretability, paving the way for safer, more reliable AI systems. Kudos to the Anthropic interpretability team for this remarkable achievement! 🎉 Link to Paper: https://lnkd.in/ehH-eCdr #AISafety #MachineLearning #AIResearch #SparseAutoencoders #Transformers #Anthropic #Claude3Sonnet #TechInnovation #ArtificialIntelligence #AIInterpretability

5

The new version of Python is out ! Check the news for f-strings in this caroussel.

13

2 commentaires

What does it take to move a machine learning project from a notebook to a real pipeline in production? Benjamin ETIENNE outlines the challenges, best practices, and practical approaches you should be aware of.

66