Lightweight Grounding¶

Grounding LLMs with minimal overhead.

Research Results¶

• Optimal weight: 95% LLM + 5% suffix array
• Perplexity reduction: 70%
• Overhead: Only 6.5% (2.66ms)

Implementation¶

See examples/lightweight_experiments.py.

This page is under construction.