# @awomanindatascience on Instagram

- **Type:** Video
- **Original URL:** https://www.instagram.com/p/DTW6O2Kk7Jj
- **Gondola URL:** https://gondola.cc/posts/61230175-awomanindatascience-instagram
- **Thumbnail:** https://img.gondola.cc/tr:w-,h-,fo-auto/postThumbnails/a192cc6266.jpg
- **Posted:** 2026-01-11T05:03:28.000+00:00
- **Account Owner:** Chinar Arora (@awomanindatascience) — https://gondola.cc/awomanindatascience

## Caption

It’s Day 13 of Building an LLM from scratch 🧠
Today’s focus: Multi-Head Self-Attention.

Instead of a single attention pass, we run attention in parallel across multiple heads, each with its own learned Q, K, V projections.
This lets the model attend to different representation subspaces and token relationships simultaneously.

Heads are concatenated (or projected) to form richer context vectors-a core reason transformers scale so well.

Implementing it from scratch makes the math click.
One block closer to a real LLM.

#deeplearning #generativeai #womenwhocode #largelanguagemodels #womenintech

## Stats

- **Views:** 3,817
- **Likes:** 341
- **Shares:** 0
- **Comments:** 6

## Tags

generativeai, womenintech, largelanguagemodels, deeplearning, womenwhocode

---
Copyright (c) Gondola