https://github.com/es7/adobe-behaviour-simulation-challenge

Problem Statements of InterIIT TechMeet 13.0 by Adobe
https://github.com/es7/adobe-behaviour-simulation-challenge

adobe artificial-intelligence interiit-techmeet machine-learning

Last synced: 9 months ago
JSON representation

Problem Statements of InterIIT TechMeet 13.0 by Adobe

• Host: GitHub
• URL: https://github.com/es7/adobe-behaviour-simulation-challenge
• Owner: ES7
• License: apache-2.0
• Created: 2024-10-11T17:53:38.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-10-17T10:51:28.000Z (over 1 year ago)
• Last Synced: 2025-08-30T09:06:52.400Z (9 months ago)
• Topics: adobe, artificial-intelligence, interiit-techmeet, machine-learning
• Language: Jupyter Notebook
• Homepage:
• Size: 1.37 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

          
# Adobe-Behaviour-Simulation-Challenge

This repository contains my team's solution for the Mid Prep Problem Statement of InterIIT TechMeet 13.0, **"Adobe Behaviour Simulation Challenge"** by Adobe.

## Data

- **1. Image_Captioning.ipynb:** In this notebook we have downloaded the media files and generate caption for them using **BLIP** model. 


- **2. 100K_Data_Generation.ipynb:** In this notebook we have reduced the length of the actual dataset (300K) and only consider the first 100K samples. 


- **3. Data_Cleaning.ipynb:** In this notebook we have combined the tokenized content text with the generated captions along with their embeddings, and then splitted this into train and test datasets.

## Task 1

- **Tokenization.ipynb:** In this notebook we have tokenized the content (captions) of the given dataset using the **DistilBERT** tokenizer. 


- **Creating_Prompts.ipynb:** In this notebook we have written prompt to convert the data into more organized format for the bucketing process. 


- **Bucketing.ipynb:** In this notebook we have created 7 buckets (classes) and trained the **DistilBERT** model for this classification task.

![](Task1/Task1_diagram.png)

## Task 2

- **FineTune.ipynb:** In this notebook we have Fine Tuned the **Bloom 7b** and **Qwen 1.5b** models using **LoRA** and **PEFT** techniques.


- **Inference.ipynb:** In this notebook, using the fine tuned models we have performed predictions on the first 500 samples.


- **Test_Metrics.ipynb:** In this notebook we have used **BLEU (1-4)**, **ROUGE (1,2,L)** and **CIDEr** metrics to evaluate the model's predictions with the original captions.


![](Task2/Task2_diagram.png)

## Team

* [**Ebad Sayed**](https://github.com/ES7) 


* [**Aaditya Jain**](https://github.com/a-aditya-jain)