Repository Overview

This repository serves as a comprehensive collection of resources and practical examples in the field of Artificial Intelligence, with a focus on Deep Learning, Natural Language Processing (NLP), and Large Language Models (LLMs).

Contents

1. PyTorch Fundamentals & Workflow (pytorch_all.ipynb)

• Tensor Operations: Demonstrations of essential PyTorch tensor manipulations including creation, reshaping, stacking, squeezing, permuting, and indexing.
• NumPy Integration: Examples of converting between PyTorch tensors and NumPy arrays.
• GPU Acceleration: How to check for and utilize CUDA-enabled GPUs for tensor operations.
• Reproducibility: Using random.seed() for consistent results.
• End-to-End PyTorch Workflow:
  • Data preparation and loading (e.g., for linear regression).
  • Splitting data into training and testing sets.
  • Building custom models using torch.nn.Module, nn.Parameter.
  • Defining forward pass.
  • Making predictions with torch.inference_mode().
  • Plotting results and model performance.

2. Hugging Face Transformers (Hugging_face_star.ipynb)

• Introduction to pipeline: Utilizing the high-level pipeline function for various NLP tasks.
  • sentiment-analysis
  • zero-shot-classification
  • text-generation (e.g., using distilgpt2)
  • fill-mask
  • ner (Named Entity Recognition)
  • question-answering
  • translation
• Transformer Architecture:
  • Conceptual overview of Encoder, Decoder, and Encoder-Decoder models.
  • Understanding the role of self-attention.
  • Use cases for different architectures (NLU vs. NLG).
• Inside the pipeline: A breakdown of the tokenization, model processing, and post-processing steps for both PyTorch and TensorFlow backends.
• Model Instantiation: Using AutoTokenizer, AutoModel, AutoModelForSequenceClassification, and AutoConfig to load and configure pretrained models.
• Transfer Learning: The significance and application of transfer learning in NLP.
• Environmental Considerations: A note on the carbon footprint of training large models.

3. Hands-on Large Language Models (LLMs) (Hands_on_LLM.ipynb)

• Text Classification:
  • Representation Models: Sentiment analysis of movie reviews (Rotten Tomatoes dataset) using transformers pipelines (e.g., cardiffnlp/twitter-roberta-base-sentiment-latest).
  • Embeddings: Using sentence transformers (e.g., sentence-transformers/all-mpnet-base-v2) for generating embeddings and performing classification.
• Generative Models & Langchain:
  • Loading Quantized Models: Implementing LlamaCpp with models like microsoft/Phi-3-mini-4k-instruct-gguf.
  • Chat Models: Interacting with ChatGoogleGenerativeAI (Gemini).
  • Agent Frameworks: Building ReAct agents with tools like DuckDuckGo search, leveraging Langchain for complex reasoning and action tasks.
• Text Clustering and Topic Modeling:
  • Applying unsupervised learning techniques to group documents (e.g., Arxiv NLP dataset abstracts).
  • Common pipeline: embeddings -> dimensionality reduction -> clustering.

4. Interactive Notebooks

• Colab Exploration (colab1.ipynb): An empty notebook, ready for interactive experiments and explorations in Google Colab.
• The repository includes Jupyter notebooks designed for hands-on learning and experimentation with the concepts mentioned above.

Getting Started

1. Clone the repository:

   git clone <repository-url>
   

2. Navigate to the desired notebook.
3. Ensure you have the necessary dependencies installed. Common libraries include torch, transformers, datasets, sklearn, sentence-transformers, langchain, llama-cpp-python, google-generativeai. You might need to install them via pip:

   pip install torch transformers datasets scikit-learn sentence-transformers langchain llama-cpp-python google-generativeai matplotlib
   

4. Open the notebooks using Jupyter Lab/Notebook or Google Colab.

Explore the contents to learn and experiment with modern machine learning and deep learning techniques.