Distributed training concepts I learned today > GPipe (Efficient training of giant neural networks using pipeline parallelism) I also made a video explaining this paper

This guy just taught FSDP on a vision pro. It can't get cooler than this.

Gemini can access your search history now. This opens up a lot of interesting prompts. It is fun to analyse yourself based on your search history.

Asked Gemini to summarise the projects I did in the last 6 months based on my search history. This is actually crazy accurate. Good work Google DeepMind

List of companies that are inventing the future of AI infra. 1) Prime Intellect 2) Nous Research 3) Pluralis Research 4) gensyn 5) EXO Labs 6) Bittensor

Day 1 of ML Journey on twitter > Studied about Clip and Siglip > Started coding a vision transformer > Revised concepts of transformers and attention

Day 2 of ML Journey > Modularized my vit implementation > Studied pali-gemma architecture in depth > Coded Layer norm, feed forward layers and general debugging

Day 3 of ML > Learnt about KV-caching in LLM's > Learnt how images and text are tokenized as input for Paligemma