Skip to content

Release 2.1.0

Choose a tag to compare

View all tags
@AlessandroPomponio AlessandroPomponio released this 05 Aug 08:11
· 139 commits to main since this release
Immutable release. Only release title and notes can be modified.
2.1.0
0819a7b
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

🏅 Highlights

  • New document resource type — ado now supports a first-class document resource, enabling structured study and report tracking directly within a project context.
  • Richer entity and space querying — entities and entity identifiers can now be fetched across multiple operations at once, with per-operation grouping, partial cache reuse, and separation of entity fetching from measurement loading.
  • Significant space-stats performance improvements — space completion checks no longer load measurement data unnecessarily, short-circuit once discrete spaces are fully sampled, and skip matchingEntities() calls when sampling is complete.
  • Expanded CLI capabilities — new flags for upgrading entities and results, filtering resources by anchor type with --related-to, and --use-latest support for datacontainers in ado describe.

✨ Features

CLI enhancements

  • ado get now supports finding related resources via a --related-to flag.
  • Space templates produced by ado now minify property domains for cleaner output.
  • A new --upgrade-entities-and-results flag on the upgrade command streamlines migrating older sample stores.
  • ado describe now accepts --use-latest for datacontainer resources.

Core improvements

  • A new document resource type is available, enabling structured notes, study tracking, and report documents to be managed as first-class ado resources.
  • Contextual error messages are now emitted when propertyDomain fails validation, making misconfigured spaces easier to debug.
  • Entity fetching and measurement result loading are now separated, giving callers finer-grained control over what data is loaded.
  • Entity identifiers and entities can be queried across multiple operations in a single call, with optional per-operation grouping.
  • Space hierarchies can now be distinguished as overlapping vs. disjoint, enabling more precise coverage analysis.

🐛 Fixes

Core

  • Experiment references in results are now derived from result data rather than a dedicated column, fixing reliability issues with cross-version data.
  • Resource models now enforce the correct kind field, preventing silent misconfiguration.
  • JSON_CONTAINS queries on MySQL and their SQLite simulation equivalent now correctly handle null value filtering.
  • Ray worker log levels are now propagated from the root logger via the runtime environment, ensuring consistent log output in remote execution.
  • The experiment catalog query is now reliable and caches results to prevent repeated I/O.
  • Accidental use of a partially populated sample store cache is now prevented with an explicit guard.

⚡ Performance

  • DiscoverySpace creation reuses a preexisting metastore connection instead of opening a new one, reducing overhead in CLI workflows.
  • Space statistics no longer load entity measurements when only counts are needed.
  • Discrete space entity matching stops as soon as the space is complete, avoiding unnecessary iteration.
  • entities_with_identifiers now partially reuses its cache across calls.

🧰 Build

  • All plugins now require ado-core>=2.0.0.
  • black has been removed from dev dependencies (superseded by ruff format).

📝 Docs

  • Agent skills have been inlined with key concepts and made self-contained.
  • CLI help text now clarifies that MEASURED_ENTITIES includes failed measurements, and that matching space similarity criteria are clearly described.
  • Documentation links have been updated to use the latest URL structure.
  • Versioned documentation publishing via mike has been added.
  • Core concepts, roadmap, and website navigation have been significantly restructured and updated.
  • Deprecated model documentation now uses nested dict utilities.
  • Agent files have been migrated from .cursor to the .agents directory.

🧹 Refactoring

  • show stats implementation has been moved alongside other show sub-commands for consistency.
  • Resource relationship lookups now use the relationship graph instead of a hierarchy, enabling more general filtering.
  • The mistyped property value workaround has been removed now that upstream data is clean.
  • Old entity fetch methods have been deprecated in favour of get_entities; upsert entity methods and their SQL statements have been removed entirely.
  • entities_in_operations has been simplified to use cache-aware helper methods.
  • Space statistics have been migrated from a table-based approach to entity-based calculations.

📈 vLLM Performance 1.13.2

  • Stale Kubernetes deployments are now garbage-collected automatically, preventing resource leaks across long-running benchmark campaigns.
  • A new prefix_repetition benchmark dataset is supported, expanding the range of workloads that can be evaluated.
  • Agent model variants are now supported in the actuator, broadening compatibility with vLLM deployments.
  • Experiment naming and versioning have been updated; older experiment names are deprecated with migration guidance.
  • Package dependencies are now correctly propagated into Ray worker runtime environments, fixing missing-dependency errors in remote execution.
  • The USER environment variable is now set to prevent uid-not-found errors on OpenShift.
  • --no-enable-prefix-caching is now correctly appended when prefix caching is disabled.

🎛️ ray-tune 2.0.7

  • A single-dimension edge case in mi_pareto_selection has been fixed, preventing a crash when the Pareto selection operates on a one-dimensional measurement space.
  • pyopenssl has been added as an explicit dependency to ensure secure connections in Ray cluster environments.
Complete changes

Complete changes

Full Changelog: 2.0.0...2.1.0

Contributors

mgazz, AlessandroPomponio, and 4 other contributors
Assets 3
Loading
0 Join discussion