TorchLeet provides a structured curriculum of PyTorch coding challenges, targeting ML engineers and researchers aiming to deepen their practical understanding of deep learning frameworks and LLM implementation. It offers a comprehensive set of problems, from basic tensor operations to advanced LLM components, with provided solutions for self-paced learning and skill development.

How It Works

The project is organized into categorized question sets (Basic, Easy, Medium, Hard) and a dedicated LLM set, each featuring incomplete code blocks (... and #TODO) within Jupyter notebooks. Users fill in the missing code to solve specific deep learning or LLM tasks, then compare their implementation against provided solution notebooks, facilitating a hands-on, problem-solving approach to learning.

Quick Start & Requirements

• Install PyTorch locally. Additional packages should be installed as needed per problem.
• Project structure includes E/M/H<ID>/qname.ipynb for problems and qname_SOLN.ipynb for solutions.
• See Getting Started for details.

Highlighted Details

• Covers a wide range of PyTorch topics: tensors, autograd, custom datasets/dataloaders, CNNs, RNNs, GANs, Transformers, and LLM components like attention, embeddings, and quantization.
• Includes specific LLM implementation challenges: RAG, speculative decoding, KV cache, LoRA, QLoRA, DPO, and continuous batching.
• Features guided learning with incomplete code and corresponding solutions for direct comparison.
• Offers problems ranging from beginner-friendly linear regression to advanced GNNs and neural style transfer.

Maintenance & Community

• Maintained by Chandrahas Aroori and Caslow Chien.
• Open to contributions for new questions or improvements.

Licensing & Compatibility

• No license is explicitly stated in the README.

Limitations & Caveats

The project does not specify a license, which may impact commercial use or redistribution. Some advanced LLM problems might require significant computational resources or specific hardware configurations not detailed in the README.