# @awomanindatascience on Instagram

- **Type:** Video
- **Original URL:** https://www.instagram.com/p/DV-Xv-mTE5R
- **Gondola URL:** https://gondola.cc/posts/61229891-awomanindatascience-instagram
- **Thumbnail:** https://img.gondola.cc/tr:w-,h-,fo-auto/postThumbnails/d33c855a70.jpg
- **Posted:** 2026-03-17T05:55:05.000+00:00
- **Account Owner:** Chinar Arora (@awomanindatascience) — https://gondola.cc/awomanindatascience

## Caption

Hello Folks! It’s Day 23 of Building an LLM from Scratch ✨

From past few weeks I have been diving into Fine tuning. Today I completed my code! It was all about instruction finetuning. Until now, my model could generate text but couldn’t reliably follow instructions. I used a dataset of ~1100 instruction–response pairs and formatted them in Alpaca style to ensure consistency. Then I tokenized the data, built custom batches, and masked padding so training focuses on meaningful tokens. Instead of training from scratch, I loaded a pretrained GPT-2 medium model and fine-tuned it for two epochs. The improvement was clear - it finally started following instructions. Biggest takeaway: pretraining teaches language, but finetuning teaches behavior.
#deeplearning #largelanguagemodels #generativeai #womenwhocode #datascience

## Stats

- **Views:** 2,095
- **Likes:** 208
- **Shares:** 0
- **Comments:** 13

## Tags

largelanguagemodels, deeplearning, womenwhocode, generativeai, datascience

---
Copyright (c) Gondola