I just created a notebook using Llama 3.1 70B to generate a synthetic data for preference finetuning / aligning LLMs: github.com/rasbt/LLMs-fro… (The updated Llama license now allows using Llama 3.1 to improve other models, yay!). You can probably guess what comes next... 😊

If you have worked with more complex PyTorch models, such as when implementing LLMs, you've probably encountered PyTorch buffers. I recorded a short 13-min hands-on coding video to explain what PyTorch buffers are, when we need them, and how we use them: youtube.com/watch?v=PetlIo…

Here's my full interview with Sebastian Raschka, one of the great AI educators today. We cover many details from DPO training failure modes, ChatGPT vs Claude, Llama 3.1, moderating Arxiv, avoiding hype, writing, getting started in AI, and other topics. Chapters: 00:00:00 Introduction &

If you want to try out the freshly released Gemma 2 2B model in Python... Just added Gemma 2 2B to LitGPT: github.com/Lightning-AI/l…

Damnnnn this tutorial is of high quality!! Check out the code repository: - Standalone Jupyter notebooks. - Only what you need. - Highly explanatory. - 0 over engineering. - All from scratch. - Self contained. I wish everyone wrote all their code like this.

I don’t post video tutorials (that) often, but hey, I just saw that I got 30k subs on YouTube! If you’re looking to learn something new this weekend, I recently made video on how LLMs work, breaking down the development stages step by step: youtube.com/watch?v=kPGTx4…

Direct Alignment Algorithms (DAA) is a great way to describe DPO et al algorithms. Endorsed. Bringing back the petition for preference fine-tuning (PrefFT) instead of RLHF).

While there are 100s of LLMs papers published each month proposing new techniques. The best way to see what truly works in practice is by looking at the pre-training and post-training pipelines of the latest state-of-the-art models. Here's what I found: magazine.sebastianraschka.com/p/new-llm-pre-…

LitServe is awesome! I also use it to serve LLMs (e.g., based on LitGPT). Short example from one of my recent talks:

If you’d like to spend a few hours this weekend to dive into Large Language Models (LLMs) and understand how they work, I've prepared a 3-hour coding workshop presentation on implementing, training, and using LLMs: youtube.com/watch?v=quh7z1…

MoEs are back!