Skip to content

Release history

deeplake releases

Deeplake is AI Data Runtime for Agents. It provides serverless postgres with a multimodal datalake, enabling scalable retrieval and training.

All releases

4 shown

v4.5.2 Maintenance

## Deeplake v4.5.2 Release Notes - Support binary data in dataset and column metadata. - Remove dependency on `libatomic`

v4.5.1 Mixed
Notable features
  • Column statistics gathering for datasets
  • Binary catalog datasets with table-catalog-type creation
  • Link content passthrough capability
Full changelog

Deeplake v4.5.1 Release Notes

Performance

  • mimalloc allocator — Switched core allocations to mimalloc for lower overhead
  • simdjson log parsing — Faster deeplog deserialization via simdjson
  • LZ4 compressed deeplog logs — Reduced on-disk deeplog size
  • String optimizations and SIMD-accelerated code paths
  • Reduced nd::array sizeof — Smaller per-array memory footprint
  • In-place downloads — Reduced memory copies in clustered index path
  • Async prefetcher priority fix

Features

  • Null value support — Full null handling in arrays, serialization, and log/replay
  • Column statistics — Statistics gathering for datasets
  • Link content passthrough — Pass content directly to link types
  • Binary catalog datasets — Table-catalog-type dataset creation

Bug Fixes

  • BM25 & inverted index — Fixed unindexed search and IN operator queries
  • exact_text search — Corrected exact text search
  • Fork safety — Fixed deadlock in async queues after fork()
  • Credentials refresh timing — Fixed refresh timing bug
  • Region detection delay — Fixed slow region auto-detection
  • Link[bytes] update — Corrected update path
  • Sanitizer fixes — Multiple UBSan/TSan/ASan issues resolved
v4.4.5 Bugfix

- Improved Deeplake support for NULL - Fixed issue with simultaneous row and column creation

Beta — feedback welcome: [email protected]