EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·27d ago·5 min read
WangNet – 1.8 MB, zero-dependency Numberwang adjudication in 11 languages
TL;DR
WangNet is a compact neural network model, just 1.8 MB in size, that determines if a number qualifies as Numberwang across 11 languages. It operates with zero dependencies, requiring only a simple Python script to run.
✦ Why It Matters
Try running WangNet today to explore its unique Numberwang adjudication capabilities without any setup hassle.
Key Takeaways
- Clone the WangNet repository to get started quickly.
- Run the model with various inputs to test its capabilities.
- Use the interactive session for a fun experience with Numberwang.
- Evaluate arithmetic expressions before inputting for better accuracy.
How It Works
WangNet reads characters directly without tokenization or normalization, using a convolutional neural network architecture to classify inputs. It leverages a simple embedding layer followed by convolutional layers and a final softmax layer to output probabilities for four verdicts: 'not Numberwang', 'THAT'S NUMBERWANG!', 'not a number', and 'Wangernumb'.