💡 [REQUEST] - Add minGRU Tutorial for Efficient Sequence Modeling

[dame-cell]

🚀 Describe the improvement or the new tutorial

I propose adding a tutorial on implementing and using minGRU (minimal Gated Recurrent Unit) to the PyTorch tutorials. This addition would provide valuable insights into efficient sequence modeling techniques for the PyTorch community.

• Efficiency: Up to 1324x faster than standard GRU for 4096-token sequences, with comparable accuracy.
• Competitive Performance: Matches state-of-the-art models like Mamba in language modeling and reinforcement learning.
• Learning Tool: Bridges simple RNNs and complex attention-based models, aiding learner progression.

Benefits for PyTorch users:

• Efficient Sequence Processing: Implement and train RNNs for long sequences, crucial for modern NLP and time series analysis.
• Parallel Training Skills: Learn to leverage parallel computing for RNN training, applicable to various deep learning tasks.
• Versatile Solution: Practical alternative to traditional RNNs and complex models, balancing efficiency and performance.

Paper

were rnns all we need

Existing tutorials on this topic

No response

Additional context

If you guys like this idea, I'm ready to jump in! I could have a PR ready as soon as tomorrow. I'm thinking of contributing a tutorial on how to use or train minGRU for language modeling

@svekars @albanD