Skip to main content

Recommended index configuration

Based on extensive testing, we recommend the following index configuration for optimal retrieval performance:

  • Chunk Size: 256 tokens
  • Chunk Overlap: 26 tokens (~10% overlap to maintain context)
  • Embedding Strategy: Semantic embedding
  • Model Name: Luminous-base
  • Representation: Asymmetric
  • Hybrid Index: BM25 (combining semantic embeddings with keyword search for comprehensive retrieval)

This recommended setup ensures efficient retrieval while preserving essential context, enhancing the overall accuracy and quality of responses provided by PhariaAssistant.