~/projects/pytorch-sota-implementations/

PyTorch SOTA Implementations

Aug 9, 2023 #pytorch#llm#deep-learning

From-scratch PyTorch implementations of state-of-the-art deep learning models, including the Transformer.

Purpose

This repository focuses on understanding modern deep-learning architectures by building them step by step in PyTorch notebooks rather than only calling ready-made models. Explanations are in English and Turkish. “SOTA” describes the subject; it does not claim current leading benchmark performance for every implementation.

Architectures

The repository includes Transformer, GPT‑1, Vision Transformer (ViT), and U‑Net implementations. Notebooks expose core layers and data flow, with research-paper PDFs and explanations for context. The collection covers designs used in text generation, image classification, and image-to-image tasks.

How to use it

English and Turkish notebooks pair code with explanations. Training settings, datasets, and evaluation vary by model; the repository is learning material for understanding how architectures are built.

← cd ..