Hello Little Coders!

If you believe AI is like Electricity, then you’d know that Electricity as such wasn’t going to create revolution without the platform to carry that - whether it’s the wires or the ground station. Deep Learning frameworks like Pytorch did that job and here’s more! 👇🏾

PyTorch Origin

PyTorch is one of the most popular Deep Learning frameworks. There were a bunch back in the day like Caffe, Theano and still some like Tensorflow, Keras, Fastai - but Pytorch has always emerged as the big winner which also seems to be a driving force this modern-day AI revolution. Here’s the origin story from Soumith Chintala 👇🏾

LLM World

1. Mixtral API pricing is continue to fall (now to zero) so that there’s are jokes about using Mixtral API and getting paid!

   

    

2. Hugging Face LLM Leaderboard Drama (Don’t read this unless you’re absolutely bored in life)

3. On a different Leaderboard, The Arena Leaderboard, Mixtral continues to climb up being the only Apache 2.0 license model at the top

   

Are you already multimodal?

If not, This is the best time to jump in. There are lot of good open models like Llava, Baklava, Qwen VL and more! This is if I ignore GPT-4V and Gemini Pro (Vision).

Here’s my tutorial on how it’s possible for you run Multimodal (Vision-Language) Model locally - even on CPU, Thanks to Llava and Ollama

A very interesting poll result about how people are turning into Multimodal models. Tbh, It’s lot more Yes than I expected!

New Datasets

1. A clean dataset is one of the most important factor for building a good model - otherwise it’s all GiGo. Thanks to Argilla’s new clean Ultrafeedback dataset which they decided to cleanup and release after their brilliant effort with Notus.

^{Subscriber Spotlight}

😟 ^{If you want, your work to be featured here, get it touch!}

Local Models

I’m always a big fan of Non-English models (even though I don’t know the language)

1. Tamil Llama - A Llama based model for Tamil Language

2. Open Hathi - An open LLM that can handle Hindi, English, Hinglish

If you build a model for your language, sure reach out to me, I’ll share it with the community!

AI Funding

1. FastAI Jeremy Howard and The Lean Startup guy launched a For-Profit (Pro Open Source) R&D Lab

2. Two Authors of the original Transformers paper have launched an Enterprise focused AI company called Essential (The last time I saw a popular figure launching Essential essentially tanked) 😉

3. a16z which is quite aggressive in their AI funding has also got a grant for Open Source related AI contributors and here’s their latest grant recipients 

Papers to read

1. Mathematical discoveries from program search with large language models

2. LLM360: Towards Fully Transparent Open-Source LLMs

3. Beyond Human Data: Scaling Self-Training for

   Problem-Solving with Language Models (from Google Deemind)

Before we end, There’s a new AI startup that’s polarized the Internet and want to know your thoughts!

What are you reading this week? Let me know!

Have feedback or interesting project, Hit reply 🙂