EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Jul 27·5 min read
How Unix spell ran in 64 kB of RAM
TL;DR
Unix engineers, led by Douglas McIlroy, ingeniously fit a 250kB dictionary into just 64kB of RAM for the Unix spell checker using advanced compression techniques.
✦ Why It Matters
Explore specialized data structures and compression techniques to optimize memory usage in your projects.
Key Takeaways
- Explore linguistics-based algorithms to reduce dictionary size.
- Implement Bloom filters for efficient membership testing.
- Use hash differences for effective data compression.
- Aim for theoretical limits in compression to optimize memory usage.
How It Works
McIlroy's approach involved a stemming algorithm to reduce words to their base forms, followed by a Bloom filter for fast lookups. As the dictionary expanded, he switched to a compressed hashing scheme that stored differences between sorted hash codes, leveraging Golomb coding for efficient compression.