Alan Buxton’s Post

It's hard to believe you can implement something like ChatGPT2 in under 1000 lines of code. I implemented ChatGPT2 in Java based on Andrej Karpathy's Llm.c. I recommend anyone interested in large language models look at llm.c. It doesn't require a GPU and will run on a decent laptop in a few minutes. Java version, the core of the model is under 900 lines. https://lnkd.in/g3rFwT2U Original Llm.c version https://lnkd.in/gj5icDyp

Do you know Foreign Function Memory (FFM)? It’s a part of the Java runtime that allows developers to directly manipulate virtual memory contents. It has been in development for the past few JDKs, and is now coming as a fully-fledged feature into JDK 22! We’ve been looking at this latest iteration, and how to do stuff with it. Moreover, the tool jextract generates Java sources from a C header, thus providing a primitive way to call functions provided by the native library. Since we’re working on embedded development at this time, we took an example that is specific to embedded development. We’re multiplying matrices on a Neural Processing Unit (a.k.a., NPU, basically a specialized processor). We’ll compare the performances with similar C++ code. https://lnkd.in/eXnx57zK

I have published more examples using Spring AI to build LLM-powered applications in Java, including image generation, speech generation, speech transcription, and function calling. https://lnkd.in/dH6QbUC4 #Java #SpringAI #LLM #GenerativeAI

I have published more examples using Spring AI to build LLM-powered applications in Java, including models from Mistral AI, Ollama, and OpenAI, and use cases like image generation, speech generation, speech transcription, and function calling. https://lnkd.in/dH6QbUC4 #Java #SpringAI #LLM #GenerativeAI

In this tutorial we will take a look at function calling in Spring AI. This will allow you to register custom Java Functions that the LLM can call when it needs access to real time data. #SpringBoot https://lnkd.in/eV_3f7R9

��🤖🤖Live recording some time ago, of writing a neural network from scratch/memory at Java level without ml libraries. It's reasonably cool/optimal to at least have fundamental neural network code in memory at least at the Java level or similar, as this can reasonably help with debugging ml libraries. 📈 📈~1000 lines of code https://lnkd.in/e4dYrRy 📝📝This is a part of a series I have practiced/teached ever since. https://lnkd.in/ej3-3w7

🚀 Modern C++: std::optional Learn how to simplify error handling and make your code more readable and consistent with the use of std::optional in C++ 17. Check out this article by Swastik Baranwal to understand the syntax, methods, and common usages of std::optional. #cpp #programming #cplusplus #coding #codenewbie Read more: [link to the article] #AI #ML #DL #ComputerVision #NLP #DataScience #BigData

It is interesting how generative A"I" shifts the attention focus from engineers producing code to engineers taking responsibility for the code they, hm, no longer "produce" but steward. Both of these have always been present, but ever more powerful (and surprisingly inept) code generators bring the responsibility aspect to the forefront. Code reading skills are indeed important for that.

check out the most used and important spring boot annotations explained in simple and easy verbose language Here's the article ,includes each annotation use case and meaning

AI code generation is about shifting the balance from writing code to reading code. You really want the code generated to be in an expressive language that makes it easy to read to verify its correctness. Readability has and will always be a key pillar of the Java language. Java for the AI world :-) #java #ai #LLM #genAI #openJDK