v1.0.0: 🎉 Fixes to demos; readme updates; cleaner release workflow

This is the first stable sieves release 🎉 No more backwards-incompatible changes going forward! v1.0.0 doesn't introduce any significant changes.

✨ New features and improvements

• Migrate all release workflow info to pyproject.toml (#249)
• Extended Python support from v3.12 to v3.12, v3.13

🔴 Bug fixes

• Fix outdated information access in demos (6cfb7ca, 1d41228)

⚠️ Backwards incompatibilities

-

📖 Documentation and examples

• Updated readme (9cc677e, 3e4043a, #252)

👥 Contributors

@rmitsch, @ivyleavedtoadflax

---

Full Changelog: v1.0.0-rc2...v1.0.0

v1.0.0-rc2: Migrate to `pyproject.toml` for releases

✨ New features and improvements

• Migrate to pyproject.toml for all release info

🔴 Bug fixes

-

⚠️ Backwards incompatibilities

-

📖 Documentation and examples

-

👥 Contributors

@rmitsch

---

Full Changelog: v1.0.0-rc1...v1.0.0-rc2

v1.0.0-rc1: Evaluation; unified result types; confidence scores; relation extraction

✨ New features and improvements

• Evaluation feature for all predictive tasks (#247)
• Uncalibrated confidence scores for all predictive tasks (#246)
• Unified predictive task result schemas to Pydantic objects (#241)
• Unified Pydantic-based prompt signature schema (#248)
• New relation extraction task (#242)
• Refactor bridges for easier addition of new task and better maintainability (#248)

🔴 Bug fixes

• Fixed streaming workflow that allocated memory for entire corpus (#248)

⚠️ Backwards incompatibilities

• Classification task requires mode: Literal[single, multi] argument instead of multi_label: bool (#244)
• Bridges have some new required methods to implement, and don't need others anymore (#248)
• Tasks have new methods to implement for evaluation and centralized task signature (#241, #248)

📖 Documentation and examples

• Case study on disaster-related tweets (#189)

👥 Contributors

@rmitsch, @adgianv

---

Full Changelog: v0.24.0...v1.0.0-rc1

v0.24.0: Better observability; clearer naming; other v1 prep

✨ New features and improvements

• Log raw model responses when include_meta is True (#236)
• Add token count tracking (#238)
• Refactor GenerationSettings to ModelSettings (#227)
• Rename Engine to ModelWrapper (#226)
• Migrate langchain support to v1
• Add single/multi mode to InformationExtraction (#233)
• Improve Pydantic model-to-HF dataset conversion (#239)

🔴 Bug fixes

-

⚠️ Backwards incompatibilities

• Refactor GenerationSettings to ModelSettings (#227)
• Rename Engine to ModelWrapper (#226)

📖 Documentation and examples

• Added links to talks and presentations (#230)
• Expanded on API docs (#234)

👥 Contributors

@rmitsch

---

Full Changelog: v0.23.0...v0.24.0

v0.23.0: Docs overhaul

✨ New features and improvements

-

🔴 Bug fixes

-

⚠️ Backwards incompatibilities

-

📖 Documentation and examples

• Overhauled documentation and readme (#225)
• Dynamic integration of test code into docs for examples to guarantee working examples in docs (#225)

👥 Contributors

@rmitsch

---

Full Changelog: v0.22.0...v0.23.0

v0.22.0: Remove optional `engines` dependency group

✨ New features and improvements

• Removes optional engines dependency group, moves all related dependencies into main group. (#210)

🔴 Bug fixes

-

⚠️ Backwards incompatibilities

• engines doesn't exist as optional dependency group anymore. When installing, instead of sieves[engines] just use sieves now. (#210)
• Dropped GLiNER2 support for sentiment analysis. (#210)

📖 Documentation and examples

-

👥 Contributors

@rmitsch

---

Full Changelog: v0.21.0...v0.22.0

v0.21.0: Unified task description handling; remove `reasoning` fields

✨ New features and improvements

• reasoning field have been removed as handling/enforcing reasoning is meanwhile better left up to the reasoning_effort configs of the underlying LM. (#205)
• Task target ID attributes (such as labels) and description attributes (such as label_descriptions) have been unified and merged into a single attribute each. E.g. labels can now be specified as Sequence[str] | dict[str, str]. (#206)
• marker uses deprecated Pydantic methods, which caused Pydantic warnings on import. Those warnings are now suppressed. (#207)

🔴 Bug fixes

• Fixes incorrect processing of GLiNER2 results for classification. (#208)

⚠️ Backwards incompatibilities

• reasoning fields are no longer supported, so any few-shot examples will have to be updated. (#205)
• Target and target description attribute (e.g. labels and label_descriptions for classification) have been merged into a target attribute that also accepts a dict. (#206)

📖 Documentation and examples

-

👥 Contributors

@rmitsch

---

Full Changelog: v0.20.0...v0.21.0

v0.20.0: Migrate to GLiNER 2

✨ New features and improvements

• Migrate from GliNER to GliNER2 (#202)

🔴 Bug fixes

• Fixes incorrect example handling in DSPy optimization (#202)

⚠️ Backwards incompatibilities

• GliNER tasks have to be updated to work with GliNER2 (#202)

📖 Documentation and examples

-

👥 Contributors

@rmitsch

---

Full Changelog: v0.19.3...v0.20.0

v0.19.3: Make progress bar optional

✨ New features and improvements

• Make progress bar optional in Pipeline (#197)

🔴 Bug fixes

-

⚠️ Backwards incompatibilities

-

📖 Documentation and examples

-

👥 Contributors

@rmitsch

---

Full Changelog: v0.19.1...v0.19.2

v0.19.1: Fix task execution on failed model prompts

✨ New features and improvements

-

🔴 Bug fixes

• Fix task support of None results generated by failed model executions (#196)

⚠️ Backwards incompatibilities

-

📖 Documentation and examples

-

👥 Contributors

@rmitsch

---

Full Changelog: v0.19.0...v0.19.1