Sitelet https://github.com/microsoft/SynapseML/tags
Skip to content

Tags: microsoft/SynapseML

Tags

v1.1.3-spark4.1

Toggle v1.1.3-spark4.1's commit message
chore: Upgrade to Spark 4.1.1, Java 17, Scala 2.13.17, Python 3.13

Spark 4.1 upgrade commit (one commit ahead of spark4.0).

Key changes:
- Spark 4.1.1, Scala 2.13.17
- Python 3.13 with updated dependencies
- langchain 0.3.0 ecosystem (langchain-community, langchain-openai)
- openai >= 1.58.1 (new API style)
- LangchainTransformer rewritten to reconstruct LLM on workers
  (avoids pickling httpx clients with thread locks)
- PEP 440 wheel filename normalization fix
- Spark 4.x API compatibility fixes

v1.1.3-python3.13

Toggle v1.1.3-python3.13's commit message
chore: Upgrade to Spark 4.1.1, Java 17, Scala 2.13.17, Python 3.13

Spark 4.1 upgrade commit (one commit ahead of spark4.0).

Key changes:
- Spark 4.1.1, Scala 2.13.17
- Python 3.13 with updated dependencies
- langchain 0.3.0 ecosystem (langchain-community, langchain-openai)
- openai >= 1.58.1 (new API style)
- LangchainTransformer rewritten to reconstruct LLM on workers
  (avoids pickling httpx clients with thread locks)
- PEP 440 wheel filename normalization fix
- Spark 4.x API compatibility fixes

v1.1.3-spark4.0

Toggle v1.1.3-spark4.0's commit message
[chore] Upgrade to Spark 4.0, Java 17, Scala 2.13, Python 3.12

Squashed from 3 previous upgrade commits:
- Changing scala version to 2.13 for spark4.0
- Upgrade to Spark4.0, Java 17, Scala 2.13, Python 3.12.11
- Spark 4.0 release bug fixes (retired AzureMaps, Language API version,
  PEP 440, flash_attn, HasReturnUsage, Fuzzing.scala)

Additional fixes during rebase:
- Scala 2.13 collection compat (getAs[Seq] → scala.collection.Seq, null guards)
- PySpark 4.0: .toDF() → createDataFrame() (removed API)
- Spark 4.0 ANSI mode: try_element_at for array indexing
- Pre-built cp312 wheels for horovod and pandas
- Fabric E2E: auto-create workspace with Runtime 2.0, dynamic Scala suffix
- DL module: try/except in __init__.py for safe jar initialization
- Lazy transformers import in HuggingFaceCausalLMTransform

v1.1.3-python3.12

Toggle v1.1.3-python3.12's commit message
[chore] Upgrade to Spark 4.0, Java 17, Scala 2.13, Python 3.12

Squashed from 3 previous upgrade commits:
- Changing scala version to 2.13 for spark4.0
- Upgrade to Spark4.0, Java 17, Scala 2.13, Python 3.12.11
- Spark 4.0 release bug fixes (retired AzureMaps, Language API version,
  PEP 440, flash_attn, HasReturnUsage, Fuzzing.scala)

Additional fixes during rebase:
- Scala 2.13 collection compat (getAs[Seq] → scala.collection.Seq, null guards)
- PySpark 4.0: .toDF() → createDataFrame() (removed API)
- Spark 4.0 ANSI mode: try_element_at for array indexing
- Pre-built cp312 wheels for horovod and pandas
- Fabric E2E: auto-create workspace with Runtime 2.0, dynamic Scala suffix
- DL module: try/except in __init__.py for safe jar initialization
- Lazy transformers import in HuggingFaceCausalLMTransform

v1.1.3

Toggle v1.1.3's commit message
fix: fix style in bump-version.py

v1.1.3-spark3.5

Toggle v1.1.3-spark3.5's commit message
fix: fix style in bump-version.py

v1.1.3-python3.11

Toggle v1.1.3-python3.11's commit message
fix: fix style in bump-version.py

v1.1.1-spark4.1

Toggle v1.1.1-spark4.1's commit message
fix: use checked-out tag for version instead of dynver (#2499)

sbt-dynver uses git-describe to derive the version, which picks
arbitrarily when multiple tags point to the same commit. This causes
builds triggered from a specific tag (e.g. v1.1.1-spark4.1) to
publish under a different tag's version (e.g. 1.1.1-python3.13).

Fix: override dynver's version with the actual checked-out tag.
- CI (ADO): uses BUILD_SOURCEBRANCH env var (refs/tags/v...)
- Local: parses git reflog to find the tag from git checkout
- Fallback: dynver (unchanged behavior for branch builds)

v1.1.1-python3.13

Toggle v1.1.1-python3.13's commit message
fix: use checked-out tag for version instead of dynver (#2499)

sbt-dynver uses git-describe to derive the version, which picks
arbitrarily when multiple tags point to the same commit. This causes
builds triggered from a specific tag (e.g. v1.1.1-spark4.1) to
publish under a different tag's version (e.g. 1.1.1-python3.13).

Fix: override dynver's version with the actual checked-out tag.
- CI (ADO): uses BUILD_SOURCEBRANCH env var (refs/tags/v...)
- Local: parses git reflog to find the tag from git checkout
- Fallback: dynver (unchanged behavior for branch builds)

v1.1.1

Toggle v1.1.1's commit message
fix: Master release bug fixes (#2485)

* fix: Master release bug fixes

- Handle retired Azure Maps Spatial API (CheckPointInPolygon throws on transform)
- Remove AzMapsPointInPolygonSuite test class and add fuzzing exemptions
- Update Language API version from 2022-10-01-preview to 2024-11-01
- Fix PEP 440 Python wheel version normalization (lowercase + local version)
- Fix conda environment activation on Linux (use conda run)

See: https://azure.microsoft.com/en-us/updates/v2/azure-maps-creator-services-retirement-on-30-september-2025

* test: Exclude geospatial notebooks from E2E tests

Azure Maps Spatial API was retired on 9/30/2025. The notebooks
'Quickstart - Flooding Risk.ipynb' and 'Geospatial Services.ipynb'
use CheckPointInPolygon which calls this retired API.

Exclude these from DatabricksCPUTests to prevent E2E test failures.