I partnered with @huggingface and @ben_burtenshaw to teach people how to fine-tune LLMs with GRPO.
In this notebook, we fine-tune a tiny SmolLM-135M model on my filtered smoltldr dataset. Thanks to our reward function, we encourage the model to output generations ("TL;DR") of around 50 characters.
Interestingly, this 135M model can learn this behavior with only 2k samples! The quality of the final results varies, but it was a fascinating exercise to tweak hyperparameters and get an intuition of how these tiny models behave with GRPO.
It's been a fun collaboration, I hope you'll enjoy the course. :)
LangChain in Action is an insightful read on building multimodal RAG applications. Roberto does a fantastic job of starting with simple archs, diving deeper into cmplx archs needed for building such applications. You can find the book here: https://t.co/vkOw6zv6bN
@ManningBooks
You are invited to the Duet AI 2024 Global Roadshow!
Join us in your city to experience the power of Duet AI and boost your development and cloud skills with #generativeAI. Register today → https://t.co/tcwCatNRXT
Really liked the content covered in Generative AI in Action: https://t.co/wfmsg4S2Rt by @bahree. The book covers the fundamentals of GenAI along with arch & code examples. Looking forward to a follow up on multimodal LLMs that are released in the future. @ManningBooks
Funny seeing adverts for Babbel and Preply - two language learning apps in the @duolingo app. Guess Duolingo is so confident people won’t jump ship that they allow competitors to advertise on their own platform 😬 #Duolingo#Competition#Advertising
This workshop will provide insights into the challenges of processing natural language, representation of words as vectors, and the significance of probabilistic and neural language models in Generative AI
https://t.co/OOUYK4WcVF
#NLP#LanguageModels#LLMs#NPTEL#workshop
1/ Thrilled to announce: Our new course ChatGPT Prompt Engineering for Developers, created together with @OpenAI, is available now for free! Access it here: https://t.co/OaIpa6L2jn
“Transformers from scratch” by Brandon Rohrer 🤖
This is one of the best write ups, that starts from 0 and explains every single detail of the model architecture.
Even if you need a refresher or don’t, I would still highly recommend reading it:
https://t.co/D25bs6TP5X
@ramaseshan Didn’t know ChatGPT wrote Tamil new year wishes but was surprised to learn it can: “புத்தாண்டு வாழ்த்துக்கள்! இந்த புது ஆரம்பம் உங்களுக்கு எப்போதும் சிறந்த விருப்பங்களை வழங்க வேண்டும்! உங்கள் வேலை மேலும் வெற்றி அடைய வேண்டும்! உங்கள் குடும்ப மேலும் எல்லை அடைய வேண்டும்!”
Our new course, "From Deep Learning Foundations to Stable Diffusion", is finally done after 8 months of work!!!
With >30 hours of video content (all free, no ads!), you'll learn how to create and train a Stable Diffusion model starting from pure Python 🧵
https://t.co/bK0PSIzFww
I asked #ChatGPT to create a 3-day itinerary to drive to Luxembourg from Germany with a dog. I was really surprised by how well it was able to create a dog-friendly travel itinerary giving notes on why it chose those places. Totally changes (yet again!) the way one plans travel!
Following Machine Learning Singapore's first back-in-person "Prompting, Instructions and the Future of Large Language Models" MeetUp, I've put together a video that explains how (for instance) #ChatGPT fits into current research directions :
https://t.co/lZfqxF0KPQ
PII or Personally identifiable information is any data that could identify an individual
🧑 Name
🏡 Address
🎂 Date of birth
💳 Credit Card number
We created a new blog on how to recognize and anonymize PII with Transformers
📕https://t.co/1EMCV8UWSf
⭐️https://t.co/f11Xo0rQHy