README.md

metadata

title: EvoTransformer V2.1
emoji: 🔥
colorFrom: pink
colorTo: blue
sdk: gradio
sdk_version: 5.36.2
app_file: app.py
pinned: false
license: apache-2.0

'''

EvoTransformer v2.1 🧠✨

EvoTransformer is an evolving neural architecture built from scratch to tackle reasoning tasks with minimal compute.

📌 What It Does

This model answers PIQA-style commonsense reasoning questions. Given a goal and two solution choices, EvoTransformer chooses the more logical one.

🔧 Architecture

• Built with 4 Transformer encoder layers
• ~13 million parameters
• Custom embedding, pooling, and classifier layers
• Fully open and adaptable for NAS or self-evolving tasks

🏋️‍♂️ Training Details

• Dataset: PIQA (1000 training, 500 validation examples)
• Optimizer: Adam
• Loss: CrossEntropy
• Epochs: 5
• Hardware: Colab GPU

🚀 Live Demo

Try it on Hugging Face Spaces: 👉 Demo Link

💡 Why EvoTransformer?

• Lean, fast, and efficient
• Custom-built from scratch (no pretraining dependencies)
• Can evolve structurally in future versions

📜 License

MIT '''