Skip to content

Feature/auto chunking#191

Open
donhardman wants to merge 2 commits into
masterfrom
feature/auto-chunking
Open

Feature/auto chunking#191
donhardman wants to merge 2 commits into
masterfrom
feature/auto-chunking

Conversation

@donhardman

Copy link
Copy Markdown
Member

No description provided.

- Implement token-accurate chunking for local models
- Add character-based heuristic chunking for remote API models
- Support configurable strategies, overlap, and maximum chunk limits
- Add utility to snap chunk boundaries to natural text separators
- Introduce ChunkSettings, ChunkSpan, and DocChunks structs
- Add make_vect_embeddings_chunked and free_chunked_result to EmbedLib
- Implement ChunkedVecResult for flat embeddings and byte-offset spans
- Export chunking functionality via FFI interface
- Fix memory leak by freeing CString error pointers in knn bindings
- Bump supported embeddings library version to 5

BREAKING CHANGE: Updated EmbedLib version and added new FFI function pointers for chunked embeddings
- Add unicode-segmentation dependency for boundary detection
- Implement UAX #29 sentence-based splitting and greedy grouping
- Introduce chunk_text dispatcher with token budgets and overlap
- Define ChunkSettings, ChunkSpan, and DocChunks for FFI compatibility
- Implement ChunkedVecResult for grouped embedding outputs
- Add token counting helpers for local and remote models
- Unify chunking logic across model implementations
- Add comprehensive tests for sentence splitting and edge cases
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant