Einstein summation lets us write an algebric expression in index-notation and have numpy/pytorch/jax generate the most effiecient set of tensor operations to compute it. The only cons is that it can be confusing and comes with a learning curve. Einsum is backend agnostic tho, i.
read more

---

Meta LLaMA, Stanford Alpaca, PaLM, OpenAI GPT-4. This week has been crazy. The previous week started with the storming LLaMA release from the Meta which advanced to weight leaks on the torrent. Then Stanford released the Alpaca model fine tuned from the LLaMA 7B model on 52K instruction-following demonstrations.
read more

---