LLM-s — Personal experiments with language & vision models

This repository is a personal collection of experiments and notebooks exploring language and vision architectures (GPT-2, BERT, ViT/DeiT, LSTM/RNN) along with a small personal chatbot project. Notebooks are intended to be readable, runnable, and educational.

Repository contents

• Notebooks (top-level):
  • bert.ipynb
  • GPT2.ipynb
  • lstm.ipynb
  • rnn.ipynb
  • Vision_transformer.ipynb
  • DeiT.ipynb
  • nano_vlm.ipynb
• Data:
  • data/MNIST/raw/ — raw IDX MNIST files for quick experiments
• Personal chatbot:
  • personal_chatbot/GPT2_project.ipynb
  • personal_chatbot/instruction-data.json
• Misc:
  • attn_is_all_u_need.png

Quick start

1. Create and activate a virtual environment (recommended):

   python -m venv .venv source .venv/bin/activate

2. Install common packages used across the notebooks (example):

   pip install --upgrade pip pip install jupyterlab notebook numpy pandas matplotlib torch torchvision transformers datasets tqdm

3. Launch Jupyter and open the notebook you want to run:

   jupyter lab

Notes:

• Many notebooks are exploratory; run them cell-by-cell and read the inline commentary.
• For training or fine-tuning, a GPU (CUDA) will significantly speed up runs. Notebooks will run on CPU but more slowly.

Personal chatbot

Open personal_chatbot/GPT2_project.ipynb for a guided example of using a GPT-2 model for a small chatbot, including tokenizer setup and fine-tuning data format (instruction-data.json).