Sometimes, nodes with long titles (more than 160 characters or so) crash the indexing process with the error: The minimum overlap cannot be equal to or exceed the maximum chunk size.
This appears to be due to the EmbeddingBase chunking both the contextual content and the main fields. When the contextual content is chunked, the chunk size is calculated using the title length. When the title is long, it can result in a calculated chunk size that is smaller than the configured minimum overlap.
Reduce the minimum overlap for contextual chunks using the Contextual content maximum percentage. This has the added benefit of significantly reducing the number of chunks generated per node. Merge request incoming.
Active
1.0
AI Search
coffeymachine → changed the visibility of the branch 3520606-minimum-overlap-error-105 to hidden.