Ryan Murray

In context of debate if large context #LLM models will eventually negate the need for #RAG, an article from Databricks has some good insights - • Longer contexts boost RAG accuracy... to a point • Performance often peaks at 32k-64k tokens, then declines • Top models (GPT-4, Claude 3.5) maintain consistency at scale • Others show unique failure modes (e.g., copyright concerns, summarization instead of Q&A) • Optimal context size varies by model and task • Lack of long-context training data may explain some issues Key takeaway: Long context and RAG are synergistic, but LLMs need refinement to fully leverage extended contexts. https://lnkd.in/eehdKtCm

17

🚀 Looking to streamline fine-tuning for models like Llama 3, Mistral, and Mixtral? Our latest blog shows how Anyscale simplifies LLMs fine-tuning with Ray: 🔑 Key takeaways: • Scalable fine-tuning with LLM-Forge • Model Deployment with RayLLM Read more: https://lnkd.in/gQwD4A2U

34

📝New article: TimeMixer - Explore the Latest Model in Time Series Forecasting Read the article here: https://lnkd.in/eatnjvXp 👈 TimeMixer proposes an MLP-based architecture that combines feature mixing with decomposition, allowing the model to capture both micro and macro variations in time series. In this article, we explore the inner workings of TimeMixer and run our own little benchmark, using Python, to evaluate the performance of TimeMixer for both short and long horizon forecasting. Enjoy the read!

27

That's awesome! Mistral's new features for building custom LLMs sound incredibly developer-friendly. The ability to fine-tune and create custom agents makes it a powerful tool for personalized AI solutions. Thanks for sharing the video link—can't wait to check it out! 🚀🔍 #CustomLLMs #AIInnovation #DeveloperTools #MachineLearning #MistralLLM

1

🚀 Microsoft has just released GraphRAG, a groundbreaking approach to Retrieval Augmented Generation (RAG) that leverages knowledge graphs to enhance LLMs' reasoning capabilities. ## Key Features of GraphRAG: - Structured Hierarchical Approach: GraphRAG extracts a knowledge graph from raw text, builds a community hierarchy, generates summaries for these communities, and leverages these structures for RAG-based tasks. - Improved Question-Answering Performance: GraphRAG shows substantial improvement in answering questions that require traversing disparate pieces of information or understanding summarized semantic concepts over large data collections. - Enhancing LLMs with Private Datasets: GraphRAG helps LLMs reason about private datasets, such as enterprise proprietary research, business documents, or communications, which the LLMs have never seen before. Getting Started with GraphRAG: - Solution Accelerator: Microsoft provides a user-friendly end-to-end experience with Azure resources to quickstart the GraphRAG system. - GitHub Repository: Check out the GraphRAG GitHub repository to access the code and documentation. - Prompt Tuning: Fine-tune your prompts following the Prompt Tuning Guide in the GraphRAG documentation for optimal results. ## Explore GraphRAG: - Microsoft Research Blog Post: Visit the Microsoft Research Blog Post to learn more about GraphRAG and how it can enhance your LLMs' abilities to reason about your private data. - Indexer and Query Packages: Dive deeper into the main sub-systems of GraphRAG by exploring the docpages for the Indexer and Query packages. Don't miss out on this exciting opportunity to leverage GraphRAG and take your LLM-based tools to the next level! 🌟 Important links in the comment. #GraphRAG #LLMs #RetrievalAugmentedGeneration #MicrosoftResearch #ArtificialIntelligence

21

1 Kommentar

Learn the power of PyTorch transpose for efficient data manipulation and model optimization! Discover how this simple operation can enhance your deep learning workflows and improve performance. Check out my article to delve into the details #PyTorch #DeepLearning #DataManipulation

3

PCA, Kernel PCA, and t-SNE: Unfolding Data’s Secrets 📊 Navigating high-dimensional data? PCA is your go-to for simplifying it—like squashing a 3D object into 2D to see the bigger picture. But what if the patterns are more complex? Enter Kernel PCA, which maps data into higher dimensions to reveal hidden structures. Now, when it comes to visualizing clusters or patterns in complex data, t-SNE is the artist 🎨. It’s like taking a tangled ball of string and laying it out so you can see the connections clearly. Together, these techniques are powerful tools for uncovering insights that might otherwise go unnoticed. Ready to dive deeper into your data? 🚀 Then reach out to us and our team would be glad to find a customised solution for your use cases! #DataScience #MachineLearning #PCA #tSNE #DataVisualization #DataAnalytics

1

"#ApacheIceberg is the most popular #opensource format for storing massive data sets and making them available to a variety of engines.   Now that the biggest #lakehouse vendors have embraced Iceberg, it’s finally possible to write data with one system and read the same data with another.  This gives developers tremendous flexibility to pick the engine of their choice for various workloads on the same data." --Dave Eyler, VP of Product at SingleStore. https://lnkd.in/eixbb3bE #BigData #DataPlatforms

18

🎉 Introducing Snowflake's groundbreaking 408B Dense + Hybrid MoE model! 🚀 🧠 17B active parameters 🤖 128 experts 📚 Trained on a massive 3.5T tokens 🔝 Uses top-2 gating for optimal performance 🆓 Apache 2.0 licensed This powerhouse excels at: ✅ SQL generation 💻 Coding 🗣️ Instruction following 🔍 4K context window Models: https://lnkd.in/dpXm9fpT

29

📣 We just released the first update to DBOS Python: v0.7.0 🚀 ! The biggest new feature is supporting 𝐪𝐮𝐞𝐮𝐞𝐬, which allows you to run many tasks (database queries, API calls, or processing jobs) concurrently. I want to highlight two key features of our 𝐏𝐨𝐬𝐭𝐠𝐫𝐞𝐬-𝐛𝐚𝐬𝐞𝐝 𝐪𝐮𝐞𝐮𝐞 implementation: 1) 𝐒𝐞𝐚𝐦𝐥𝐞𝐬𝐬 𝐬𝐜𝐚𝐥𝐚𝐛𝐢𝐥𝐢𝐭𝐲: Tasks can be distributed across nodes without needing to set up an external orchestrator or state store, as DBOS uses your app's database as both the orchestrator and queue. To scale, simply launch additional instances of your app. 2) 𝐒𝐢𝐦𝐩𝐥𝐢𝐟𝐢𝐞𝐝 𝐢𝐧-𝐨𝐫𝐝𝐞𝐫 𝐩𝐫𝐨𝐜𝐞𝐬𝐬𝐢𝐧𝐠 𝐚𝐧𝐝 𝐜𝐨𝐧𝐜𝐮𝐫𝐫𝐞𝐧𝐜𝐲 𝐥𝐢𝐦𝐢𝐭𝐬: For example, we now support exactly-once, in-order processing for Kafka events. This is also helpful for controlling the number of concurrent requests to external services, such as LLMs, to avoid overloading downstream services. Some other cool features in this release include the support for workflows publishing "status updates" through mutable keys in the events API, and upgrading to psycopg3. We are on our way to support 𝐀𝐬𝐲𝐧𝐜𝐢𝐨, stay tuned! Full release notes: https://lnkd.in/djESH8dS

23

🔥 SQLite-Vec == VectorDB + OnDevice + SQLite - small memory footprint - written in C - no dependencies pip install sqlite-vec 🎁 GitHub: https://lnkd.in/gRSSEgJe 📚 Docs: https://lnkd.in/gsxGqFCF 🎉 Blog: https://lnkd.in/gJDUYFJw 👉 What is SQLite-Vec: An extremely small, "fast enough" vector search SQLite extension that runs anywhere! A successor to sqlite-vss #database #sql #vectordb #ml #python #c #pylib #sqlitevec #sqlite #datascience

13

Growth with Custom Neural Machine Translation (NMT) & LSPs Precise Translation, Localization #L10N, and Internationalisation #I18N for your software, knowledge bases, training and documentation is essential for international success. Why NMT? Both advanced NMT models and LLMs (GPTs) leverage the state-of-the-art Transformer Architecture. NMTs specialise in translation tasks, resulting in superior accuracy for industry specific e.g. finance, legal, and technical domains. They have been arround for longer then LLMs. NMTs are the engine behind most of the AI translation tools and workflows, such as: ♻️ Hybrid Translation 1. Machine Translation (MT): NMT engine performs the initial translation. 2. Quality Prediction: Tools like ModelFront can analyse the MT output and predict segments with high quality that might not require human editing. 3. Human Post-Editing: Translators focus on segments identified as needing improvement or specific domain expertise. 4. Terminology Management: Some TMS and CAT tools ensure advanced jargon consistency. 5. Machine Learning #ML: Improve 1. MT, ⚠️ consider the ownership of your Translation Memories and Privacy ☑️ The Challenge: Your Custom NMT ✔️ Your options: ✅️ Partner with Language Service Provider (LSP) with Established NMT capabilities: Google Translate | Amazon Translate Microsoft Translator | IBM Watson Translator Lionbridge | XTM Cloud | DeepL WorldServer, Language Weaver RWS | SDL Crowdin | Smartling | Translated | LanguageWire ✅️ Develop - Experiment - Learn In-House (open source) #Python: #OpenNMT ⚡️Sockeye https://lnkd.in/eumhW6DW Fairseq by AI at Meta ⚠️ C++ core T5 by Google Research AI ✔️ Benefits: Superior Accuracy: User-centric, can be critical Integration: Language Services with your workflows and strategy Domain-Specific Terminology Management and Transcreation 👏 Your Expertise Matters! We'd love to hear from practitioners who have tackled these challenges firsthand. Share your experience with custom NMT in the comments! 👏 Have I missed an interesting NMT open source, affordable developers ressource or a #LSP? #Fintech #Regtech #NeuralMachineTranslation #LLM #NLP #AIProductManagement #AITranslation #ProductGrowth

25

24 Kommentare

https://lnkd.in/dBkJGXwN A really detailed paper comparing latest results in linear mat-mul free progress. Worth reading and if you are a Ph.D. Researcher something to consider as a great investment for your carreer

6

Short video on a great paper - tree based models out performing neural nets on tabular data (one could stretch this to say structured and semi-structured data) https://lnkd.in/gqfZdsmX Neural Nets are great for unstructured data… nevertheless, message should be clear - right tool for the right problem

6

2 Kommentare

👉 Housekeeping in Embedding Models & Vector DBs … is crucial and as important as selecting the right vector model. Great insights from MongoDB‘s Philip Eschenbacher The next big thing will be hybrid models from the team to move the needle for GenAI Live from GenAI Zürich #GenAIZurich #GenAIZurich2024

17

3 Kommentare