Paul Klinger

How often does an opportunity come to work on a greenfield platform like Pydantic Logfire? Build a database that will scale (if we get it right) to petabyte scale and change how develoeprs understand their applications.

115

3 Comments

I have scaled up the popular UCI Machine Learning wine quality dataset from 178 rows to an awesome 100k rows. This was accomplished using YData's Python module, with an accuracy of 90%. Data scientists and machine learning practitioners now have more rows of the wine quality dataset with which to train their classification models for enhanced accuracy. Download the dataset here: https://lnkd.in/dA-6UJ7m Have a dataset you want to scale up? Reach out! 👍 machinelearning.fr Machine Learning Machine Learning Mastery Machine Learning Machine Learning Labs Joachim Schork

2

Working on some personal AI projects tonight (will share a GitHub repo soon). I’ve spent a few hours over several days in one specific chat with o1-preview. I asked it a particularly interesting question, and you should too. It’s great to have an AI sidekick to help get work done and be super supportive! #AI #QuestionsToAsk #Automation #PowerOfReflection #ChainOfThought

3

I wrote a short tutorial on using Porcupine to check for linearizability (without needing to deal with the JVM). https://lnkd.in/eW6Y-QQQ

64

“Explain options trading in simple terms” while I’m on the tube with no internet 🛜 Here’s a very awkward demo on the London Underground of me using #opensource AI tooling (Ollama and Phi-3) to converse in natural language without an internet connection. Why is this important? As models get smaller, our devices get more powerful and all this compute happens locally (your devices - computers, phones, toasters?) we will begin to see new ways to work with models in new places. Here an example: A education application that could be used on your commute that tailors itself to you as you go. Currently that requires internet but with local on device models you could that anywhere, anytime and for free* Phi-3 is “tiny” and could fit on a phone. Its been done before. Apple and Google want to control the model on our phones but we shall see if this actually happens. An OpenAI phone? Maybe but I doubt it. Anyways just thought I’d try this and share it here for impact. We are moving towards on device AI compute. What a time to be alive.

18

5 Comments

The final part of my Apache Iceberg consistency model series is out, covering the formal verification work. It's also the end of the table format consistency model series, but not the end of my writing about the table formats. More still to come... https://lnkd.in/e2hrPKKZ

37

Pydantic is sponsoring Frank Hoffmann because inline-snapshots is transformative to how we write unit tests in Python! https://lnkd.in/en6HtXCF If you're using pytest, and you're not using inline-snapshots, you're wasting your time; simple as that.

335

19 Comments

This is a TIL (well, not quite today) post in relation to Spark Scala. I spent way too much time finding what was causing a race condition in my code a few days ago, and fixed it by using foldLeft instead of map. I then noticed something interesting - a frankly speaking outrageous performance improvement of 30%, so I did a little bit of digging. Turns out, single-threading certain bits of code can make it more performant. Here are the function signatures: // map function signature: def map[B](f: A => B): Dataset[B] // foldLeft function signature def foldLeft[B](z: B)(op: (B, A) => B): B - **map Function:** Inherently parallelisable, but creates intermediate collections. - **foldLeft Function:** Inherently single-threaded, processes elements sequentially, without generating intermediate collections. The performance improvement seems to come from the fact that foldLeft does not create intermediate collections. This means that: (1) within a large number of groups, especially when operating within a keyValueGroupedDataset, there's a ton of memory overhead when using map; (2) we save on the execution steps of memory allocation and garbage collection; and (3) with a single accumulator, we can capitalise on cache locality, and are less likely to get cache misses like we would if it did not fit on the CPU cache. I did not expect such a significant improvement in performance, and will definitely be using foldLeft more often in the future.

9

GraphRAG in a few sentences for the busy (This is at a very high level and not exhaustive coverage) - 1. Opensource Microsoft framework. 2.  a Python library for extracting insights from unstructured textual data and representing them as graphs using LLMs. 3. MIT licence 4. Release link - https://lnkd.in/e_ukifCc 5 a. Git link - https://lnkd.in/e-qtfqgV 5 b. https://lnkd.in/eJW9_yMR P.S.- I need to play around more to get more insight into it. Feel free to let me know if you guys have already tried this. #graph #ontologies #llm #datascience #machinelearning

5

🌟 Pioneering Advanced Retrieval Techniques for RAG Applications 🌟 Hi LinkedIn community, In my journey to enhance Retrieval-Augmented Generation (RAG) systems, I’ve been implementing and experimenting with advanced retrieval techniques. So far, I’ve worked on: HyDE (Hypothetical Document Embeddings) Vector Search RRF (Reciprocal Rank Fusion) Fusion Retrieval To share my findings and progress, I’ve created a GitHub repository: 🔗 https://lnkd.in/gwAeYjTK This is a growing collection of tools, implementations, and insights into how retrieval techniques can make RAG systems smarter, faster, and more scalable. 💡 If you’re exploring similar areas or have insights to share, I’d love to collaborate! Let’s work together to push the boundaries of retrieval and generative AI. #RetrievalAugmentedGeneration #AI #AdvancedRetrieval #InformationRetrieval #RetrievalTechniques

14

2 Comments

My latest research work, "Towards smaller, faster decoder-only transformers: Architectural variants and their implications" is now available on arXiv. In this work, we introduce architectural variations to the decoder-only transformer architecture for easier deployment, faster training and faster inferencing with little drop in performance. Do check it out in the link below and feel free to share your thoughts and insights. https://lnkd.in/gF_A4HeU #research #machinelearning #deeplearning

51

2 Comments

What if your OpenAI API calls were 100% reliable? This is the claim from the Structured Outputs announcement yesterday. Here is the excerpt: "On our evals of complex JSON schema following, our new model gpt-4o-2024-08-06 with Structured Outputs scores a perfect 100%. In comparison, gpt-4-0613 scores less than 40%." I have been using libraries like instructor to get structured output, but this might mean that all you need is the API now. What do you think? Link: https://buff.ly/46SD2Cd

2

𝗘𝘅𝗰𝗶𝘁𝗶𝗻𝗴 𝗡𝗲𝘄𝘀 𝘄𝗶𝘁𝗵𝗶𝗻 𝘁𝗵𝗲 𝗟𝗮𝗿𝗴𝗲 𝗟𝗮𝗻𝗴𝘂𝗮𝗴𝗲 𝗠𝗼𝗱𝗲𝗹𝘀 (𝗟𝗟𝗠𝘀) 𝗦𝗽𝗮𝗰𝗲! PyTorch has officially released TorchTune, a new tool designed to fine-tune Large Language Models (LLMs) with ease and efficiency. This groundbreaking tool promises to streamline the fine-tuning process, making it more accessible to researchers and practitioners alike. https://lnkd.in/eMf6dNZJ 𝗧𝗼𝗿𝗰𝗵𝘁𝘂𝗻𝗲 𝗢𝗳𝗳𝗲𝗿𝘀: ▪ 𝗣𝘆𝗧𝗼𝗿𝗰𝗵-𝗻𝗮𝘁𝗶𝘃𝗲 𝗟𝗟𝗠 𝗶𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻𝘀: Build on familiar ground with composable building blocks for popular LLMs. ️ ▪ 𝗠𝗲𝗺𝗼𝗿𝘆-𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗶𝗻𝗴 𝗿𝗲𝗰𝗶𝗽𝗲𝘀: Train on consumer-grade GPUs with 24GB VRAM – no beefy machines needed! ▪ 𝗘𝘅𝘁𝗲𝗻𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆 𝗮𝗻𝗱 𝗲𝗮𝘀𝗲 𝗼𝗳 𝘂𝘀𝗲:  YAML configs and support for common dataset formats streamline your workflow. 𝗙𝗼𝗿 𝘁𝗵𝗼𝘀𝗲 𝗶𝗻𝘁𝗲𝗿𝗲𝘀𝘁𝗲𝗱 𝗶𝗻 𝗱𝗶𝘃𝗶𝗻𝗴 𝗱𝗲𝗲𝗽𝗲𝗿, 𝗵𝗲𝗿𝗲 𝗮𝗿𝗲 𝘀𝗼𝗺𝗲 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀: Code: https://lnkd.in/gEgETe5t Blog: https://lnkd.in/gP3mctA6 Tutorials: https://lnkd.in/gvh2eEsU This is a fantastic opportunity to explore and innovate, I'd be very interested in hearing your thoughts. ▪ 𝗪𝗵𝗮𝘁’𝘀 𝘆𝗼𝘂𝗿 𝘁𝗮𝗸𝗲 𝗼𝗻 𝘁𝗵𝗶𝘀?  ▪ 𝗛𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝘀𝗲𝗲 𝗧𝗼𝗿𝗰𝗵𝗧𝘂𝗻𝗲 𝗶𝗺𝗽𝗮𝗰𝘁𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝗰𝘂𝗿𝗿𝗲𝗻𝘁 𝗽𝗿𝗼𝗷𝗲𝗰𝘁𝘀 𝗼𝗿 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵? ▪ 𝗛𝗼𝘄 𝗱𝗼 𝘆𝗼𝘂 𝘁𝗵𝗶𝗻𝗸 𝗧𝗼𝗿𝗰𝗵𝗧𝘂𝗻𝗲 𝘄𝗶𝗹𝗹 𝗶𝗺𝗽𝗮𝗰𝘁 𝘁𝗵𝗲 𝗳𝘂𝘁𝘂𝗿𝗲 𝗼𝗳 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗮𝗻𝗱 𝗔𝗜 𝗱𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁? #AI #PyTorch #TorchTune #LLMs #MachineLearning

24

1 Comment

We've all wrestled with messy data like the one below - tables trapped in PDFs, screenshots with merged cells, and nested structures that break our matplotlib code. With gpt-4o and instructor, we can now go from messy screenshots by automatically processing table images into tidy pandas dataframes. The secret here? Getting our model to output a markdown table that we validate using pandas with a custom data type. This allows us to breakdown a complex nested table into multiple smaller dataframes for our specific use case and further analysis. Check out our latest article on how to implement this at https://lnkd.in/gFcJ7y3y

3

Here's a good read - an article I recently wrote about epochs and how to tune them: Tuning Epochs: Because Your Model's Been Lazy Long Enough #machinelearning #epoch

2

How do you evaluate a generative AI system when there’s an unbounded set of right answers? Check out our latest blog post where we gave the LLM a gavel 🧑‍⚖

6

This week I have been thinking about security models with generative AI. Have a read and let me know your thoughts in the comments section! #MethodsForGood #ai #generativeai #security #cybersecurity #responsibleai# #dataprotection

9