DL-paper_implementations

A curated collection of deep learning paper implementations built only using NumPy and PyTorch.
This repo spans from classic architectures to modern transformer-based models, along with experiments in fine-tuning and representation learning.

✅ Implemented Papers & Architectures

• Foundational Architectures

  • LeNet
  • AlexNet

• Representation Learning and Sequence Modeling

  • Vanilla RNNs
  • LSTMs
  • word2vec
  • GloVe
  • Learning Phrase Representations using RNN Encoder–Decoder

• Transformer Architectures

  • BERT
  • GPT-1
    • GPT-1 (pytorch) fine-tuned for generating Shakespeare-like stories
  • GPT-2 (adapted for various tasks)
  • Llama-2

• Neural Machine Translation

  • investigating multilingual nmt representations at scale

• Fine-Tuning Techniques

  • LoRA (Low-Rank Adaptation)

• Generative Models

  • GANs
  • DCGans
  • WGans
  • WGan-gp

• Vision

  • Vision Transformers

• metrics

  • bleu

• finetuning

  • classification finetuning gpt2 from scratch
  • instruction finetuning

---

🧠 All models are implemented from scratch or with minimal reliance on high-level libraries, to deepen understanding of core concepts and techniques.