dspy-ai

What's Changed

• Switch settings from contextvar to thread local storage (for Colab) by @okhat in https://github.com/stanfordnlp/dspy/pull/1860
• Signature.replace compatibility with pydantic 2.10 by @mikeedjones in https://github.com/stanfordnlp/dspy/pull/1855
• Allow react.Tool to wrap methods by @tkellogg in https://github.com/stanfordnlp/dspy/pull/1856
• Adapters: Support JSON serialization of all types via pydantic (e.g. datetimes, enums, etc.) by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1853

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.39...2.5.40

fix settings contextmanager

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.38...2.5.39

• add back settings.config

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.37...2.5.38

Hotfix over 2.5.36

• Fix settings.context finally block: main_thread_config

• fix settings.py

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.36...2.5.37

What's Changed

• Convert dspy.settings to a ContextVar
• Improve ParallelExecutor (isolate even if 1 thread)
• Permit user-launched threads

by @okhat in https://github.com/stanfordnlp/dspy/pull/1852

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.35...2.5.36

What's Changed

• Expand default cache limit to 30GB by @okhat in https://github.com/stanfordnlp/dspy/pull/1850
• Update RAG tutorial to use the downsampled RAG-QA Arena upload by @okhat in https://github.com/stanfordnlp/dspy/pull/1851

New Contributors

• @b-d055 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1848

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.34...2.5.35

What's Changed

• Introduce dspy.retrievers.Embeddings index (thread-safe, memory-friendly faiss kNN)
• Introduce CompleteAndGrounded metric
• Introduce Unbatchify utils

by @okhat in https://github.com/stanfordnlp/dspy/pull/1843

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.33...2.5.34

What's Changed

• Add teacher module to MIPROv2 by @Shangyint in https://github.com/stanfordnlp/dspy/pull/1830
• Make litellm version more flexible by @rogeriochaves in https://github.com/stanfordnlp/dspy/pull/1833
• Updating MIPROv2 Logging by @klopsahlong in https://github.com/stanfordnlp/dspy/pull/1828
• Style fix for Dataloader by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1838
• Fix dspy.Evaluate's handling of exceptions (from 2.5.30) by @okhat in https://github.com/stanfordnlp/dspy/pull/1839

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.32...2.5.33

What's Changed

• Revert back to more permissive dspy.settings multi-threaded access by @okhat in https://github.com/stanfordnlp/dspy/commit/9a70c7c42c
• Fix additional evaluation display failures for non-dict objects by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1820
• Use dspy.Embedding for KNN by @CyrusOfEden in https://github.com/stanfordnlp/dspy/pull/1822
• Replace print by logger.info in BetterTogether optimizer by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1827

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.31...2.5.32

What's Changed

• Fix asyncify but warn on settings read from unrecognized thread by @okhat in https://github.com/stanfordnlp/dspy/pull/1813

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.30...2.5.31

What's Changed

• Change the way of enable/disable litellm logging by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1777
• Litellm dependency by @Florents-Tselai in https://github.com/stanfordnlp/dspy/pull/1779
• fix: prevent Chinese examples from being converted to Unicode encoding by @coolmian in https://github.com/stanfordnlp/dspy/pull/1787
• Dev finetune touches by @dilarasoylu in https://github.com/stanfordnlp/dspy/pull/1796
• (Feature) dspy.asyncify by @CyrusOfEden in https://github.com/stanfordnlp/dspy/pull/1806
• fix: for logging to console w image urls by @glesperance in https://github.com/stanfordnlp/dspy/pull/1794
• Add support for native parallel execution in DSPy by @krypticmouse in https://github.com/stanfordnlp/dspy/pull/1690
• Make sure threading fails without ParallelExecutor fails (if settings are accessed) by @okhat in https://github.com/stanfordnlp/dspy/pull/1810
• Generating reliability test programs and inputs by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1802 and https://github.com/stanfordnlp/dspy/pull/1808

New Contributors

• @Florents-Tselai made their first contribution in https://github.com/stanfordnlp/dspy/pull/1779
• @glesperance made their first contribution in https://github.com/stanfordnlp/dspy/pull/1794

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.29...2.5.30

What's Changed

• Databricks finetuning integration by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1770
• fix: prevent Chinese examples from being converted to Unicode encoding by @coolmian in https://github.com/stanfordnlp/dspy/pull/1774

New Contributors

• @coolmian made their first contribution in https://github.com/stanfordnlp/dspy/pull/1774

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.28...2.5.29

What's Changed

• Revamp BootstrapFinetune and merge BetterTogether to main by @dilarasoylu in https://github.com/stanfordnlp/dspy/pull/1698
• Add test coverage for caching against a LiteLLM test server by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1769
• add flag to suppress litellm logs in dspy.LM by @arnavsinghvi11 in https://github.com/stanfordnlp/dspy/pull/1768

New Contributors

• @kingtroga made their first contribution in https://github.com/stanfordnlp/dspy/pull/1653

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.27...2.5.28

What's Changed

• Upgrade pinned litellm to 1.51.0
• Revert LiteLLM Router-based retries and upgrade poetry lock for litellm 1.51.0 by @okhat in https://github.com/stanfordnlp/dspy/pull/1762

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.26...2.5.27

What's Changed

• Remove structlog dependency by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1751
• LMs: retry with exponential backoff for a limited set of error codes by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1753
• Fixes for OpenAI / Azure OpenAI compatibility with LiteLLM router by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1760
• Fixed by exluding attestations which were automatically enabled in an… by @hmoazam in https://github.com/stanfordnlp/dspy/pull/1761

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.25...2.5.26

What's Changed

• Updates to dspy.ReAct by @okhat in https://github.com/stanfordnlp/dspy/pull/1716

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.19...2.5.20

What's Changed

• Revamp ReAct, adjust Bootstrap, adjust ChatAdapter by @okhat in https://github.com/stanfordnlp/dspy/pull/1713

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.18...2.5.19

What's Changed

• JsonAdapter: Handle JSON formatting in demo's outputs by @okhat in https://github.com/stanfordnlp/dspy/pull/1702

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.17...2.5.18

What's Changed

• Deprecate dspy.TypedPredictor and its variants. You'll get much better typed (and untyped) results out-of-the-box now with standard dspy.Predict, dspy.ChainOfThought, etc.
• Improve ChatAdapter, introduce JsonAdapter, add default retries with the latter. by @okhat in https://github.com/stanfordnlp/dspy/pull/1700
• Consistently use _parse_values=False from functional.py by @okhat in https://github.com/stanfordnlp/dspy/pull/1701
• Add callback mechanism by @B-Step62 in https://github.com/stanfordnlp/dspy/pull/1689
• Some minor fixes to the newly introduced callback by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1696

New Contributors

• @BobConanDev made their first contribution in https://github.com/stanfordnlp/dspy/pull/1688
• @B-Step62 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1689

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.16...2.5.17

What's Changed

• Anyscale finetuning refactor by @isaacbmiller in https://github.com/stanfordnlp/dspy/pull/1666
• Fix error in Evaluate with display_table=True with outputs that cannot be converted to dict by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1682

New Contributors

• @amanshresthaa made their first contribution in https://github.com/stanfordnlp/dspy/pull/1667
• @Haknt made their first contribution in https://github.com/stanfordnlp/dspy/pull/1671
• @sushmanthreddy made their first contribution in https://github.com/stanfordnlp/dspy/pull/1677

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.15...2.5.16

What's Changed

• Improve ChatAdapter's handling of typed values and Pydantic models by @okhat in https://github.com/stanfordnlp/dspy/pull/1663
• Push Updated Docs view based on MkDocs by @krypticmouse in https://github.com/stanfordnlp/dspy/pull/1657

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.14...2.5.15

What's Changed

• Fix TypedPredictor formatting with list output values by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1609

New Contributors

• @YangZhaoo made their first contribution in https://github.com/stanfordnlp/dspy/pull/1652

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.13...2.5.14

What's Changed

• Refactor finetuning implementation to be 2.5 compatible by @isaacbmiller @dilarasoylu in https://github.com/stanfordnlp/dspy/pull/1594
• Fix the saving of TypePredictor by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1651

New Contributors

• @dilarasoylu made their first contribution in https://github.com/stanfordnlp/dspy/pull/1633

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.12...2.5.13

What's Changed

• Fix saving/loading retriever by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1638

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.11...2.5.12

Add SemanticF1, fix LM.copy, fix CoT.save

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.10...2.5.11

What's Changed

• Move path creation to BSFS init by @isaacbmiller in https://github.com/stanfordnlp/dspy/pull/1632
• Fix: 'LM' object has no attribute 'copy' by @xiayouran in https://github.com/stanfordnlp/dspy/pull/1621
• support TEI by @xiayouran in https://github.com/stanfordnlp/dspy/pull/1553

New Contributors

• @xiayouran made their first contribution in https://github.com/stanfordnlp/dspy/pull/1621

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.9...2.5.10

What's Changed

• add support for reasoning/rationale in MultiChainComparison by @arnavsinghvi11 in https://github.com/stanfordnlp/dspy/pull/1630

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.8...2.5.9

What's Changed

• Patch Poetry build fail due to litellm.caching by @krypticmouse in https://github.com/stanfordnlp/dspy/pull/1629

New Contributors

• @yujonglee made their first contribution in https://github.com/stanfordnlp/dspy/pull/1623

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.7...2.5.8

What's Changed

• MIPROv2 Refactor & Documentation Updates by @klopsahlong in https://github.com/stanfordnlp/dspy/pull/1616
• Other smaller updates

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.6...2.5.7

What's Changed

• Fix signature saving at Predict saving by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1556

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.5...2.5.6

What's Changed

• New release flow with dspy-ai as a wrapper depending on dspy and dspy… by @hmoazam in https://github.com/stanfordnlp/dspy/pull/1584

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.4...2.5.5

What's Changed

• Add num_retries to signature_opt_typed by @tkellogg in https://github.com/stanfordnlp/dspy/pull/1568
• Fix MIPROv2 exception on demoless runs by @XenonMolecule in https://github.com/stanfordnlp/dspy/pull/1581
• [MIPROv2] Fixed min to max in demo check by @XenonMolecule in https://github.com/stanfordnlp/dspy/pull/1591
• updated assertions error msg to specify target_module by @arnavsinghvi11 in https://github.com/stanfordnlp/dspy/pull/1590
• LangProbe Benchmark Updates by @klopsahlong in https://github.com/stanfordnlp/dspy/pull/1593
• chat_adapter: Format pydantic fields as JSON by @tkellogg in https://github.com/stanfordnlp/dspy/pull/1569
• Enhance history with filterable content by @Anindyadeep in https://github.com/stanfordnlp/dspy/pull/1562
• Remove get_convo asserts from tests by @mikeedjones in https://github.com/stanfordnlp/dspy/pull/1595
• Update docs for LMs and migrations by @krypticmouse in https://github.com/stanfordnlp/dspy/pull/1587
• Update faq docs with new cache info by @krypticmouse in https://github.com/stanfordnlp/dspy/pull/1602

New Contributors

• @tkellogg made their first contribution in https://github.com/stanfordnlp/dspy/pull/1571
• @kevin-v96 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1583

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.3...2.5.4

What's Changed

• fix(dspy): bug fix for snowflake retriever caused by smart filter update by @sfc-gh-alherrera in https://github.com/stanfordnlp/dspy/pull/1565
• [1] DatabricksRM: Use databricks-sdk to fetch token / workspace URL + several small improvements by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1564

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.2...2.5.3

What's Changed

• Fixing typed predictors for 2.5 with a temporary _parse_values=False … by @okhat in https://github.com/stanfordnlp/dspy/pull/1558
• Improve o1 support in dspy.LM (QoL errors) by @okhat in https://github.com/stanfordnlp/dspy/pull/1559

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.1...2.5.2

What's Changed

• Make output processing in typed predictors compatible to dspy.LM (2.5) by @krypticmouse in https://github.com/stanfordnlp/dspy/pull/1546
• Patche python interpreter to support additional syntax by @arnavsinghvi11 in https://github.com/stanfordnlp/dspy/pull/1550
• Release workflow updated to also publish to dspy as well as dspy-ai on pypi by @hmoazam in https://github.com/stanfordnlp/dspy/pull/1557

New Contributors

• @bluewin4 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1542
• @jannikmaierhoefer made their first contribution in https://github.com/stanfordnlp/dspy/pull/1554

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.5.0...2.5.1

What's Changed

• Adapters type casting & enforcement. Migration notebook. by @okhat in https://github.com/stanfordnlp/dspy/pull/1532
• Add deprecation warning for old clients on first use (per client) by @okhat in https://github.com/stanfordnlp/dspy/pull/1533
• Pin cached notebooks to 2.4.17 prior to deprecations by @okhat in https://github.com/stanfordnlp/dspy/pull/1528
• Top-level file cleanup by @okhat in https://github.com/stanfordnlp/dspy/pull/1529
• Add warning for deep copy by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1531

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.4.17...2.5.0

What's Changed

• Various improvements for adapters & LMs by @okhat in https://github.com/stanfordnlp/dspy/pull/1523
• MIPROv2(v2) -- Big Updates to the MIPROv2 Optimizer by @XenonMolecule in https://github.com/stanfordnlp/dspy/pull/1522
• Improve error logging with a stacktrace by @severi in https://github.com/stanfordnlp/dspy/pull/1458
• Fix dspy Module's deepcopy by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1519
• Typo/nit fixes and add tests for BoostrapWithRandomSearch by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1502

New Contributors

• @siddharthsambharia-portkey made their first contribution in https://github.com/stanfordnlp/dspy/pull/1496
• @NirantK made their first contribution in https://github.com/stanfordnlp/dspy/pull/1498
• @PrashantDixit0 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1444
• @jeffreyrubi made their first contribution in https://github.com/stanfordnlp/dspy/pull/1484
• @manubhardwaj made their first contribution in https://github.com/stanfordnlp/dspy/pull/1437
• @arya-krishna-23 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1432
• @severi made their first contribution in https://github.com/stanfordnlp/dspy/pull/1458

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.4.16...2.4.17

This is a pretty significant release. New dspy.LM and dspy.Adapter will be key for DSPy 2.5 onwards. Old clients are unaffected for now.

What's Changed

• Adding new dspy.LM and dspy.Adapter by @okhat in https://github.com/stanfordnlp/dspy/pull/1486
• Make langchain copy-able in DSPy by @chenmoneygithub in https://github.com/stanfordnlp/dspy/pull/1488
• feat(dspy): support o1 parameters by @isaacbmiller in https://github.com/stanfordnlp/dspy/pull/1487

New Contributors

• @rawsh made their first contribution in https://github.com/stanfordnlp/dspy/pull/1485

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.4.15...2.4.16

What's Changed

• Assertions patches by @arnavsinghvi11 in https://github.com/stanfordnlp/dspy/pull/1471
• Fix misc assertion errors by @Shangyint in https://github.com/stanfordnlp/dspy/pull/1085
• Fix assertion bug with new_signature by @Shangyint in https://github.com/stanfordnlp/dspy/pull/1372
• Fix rendering of evaluation results Pandas DataFrames in console environments by @dbczumar in https://github.com/stanfordnlp/dspy/pull/1472
• Bug Fix: Multiline init regex parsing (MIPROv2) by @XenonMolecule in https://github.com/stanfordnlp/dspy/pull/1466

New Contributors

• @karthikscale3 made their first contribution in https://github.com/stanfordnlp/dspy/pull/1456
• @hacker-4-good made their first contribution in https://github.com/stanfordnlp/dspy/pull/1469
• @dbczumar made their first contribution in https://github.com/stanfordnlp/dspy/pull/1472

Full Changelog: https://github.com/stanfordnlp/dspy/compare/2.4.14...2.4.15

• supports configurable backoff time for LM/RM providers dspy.settings.configure(backoff_time = ...)
• latest PR changes and updates (Langchain Metaclass resolution, addition of Langchain Tool Execution, MIPROv2 parameter updates, LM/RM Provider updates)

Full Changelog: https://github.com/stanfordnlp/dspy/compare/v2.4.12...2.4.13

• Supports treating dspy.Predict and dspy.ChainOfThought directly as Modules, e.g. can compile them without creating a wrapping dspy.Module object.

• Improves the experimental=True Chat LM support from DSPy v2.4.11, which is quoted below for reference:

Initial support for new adapters, with far more smooth support for Chat LMs. The default behavior is unchanged from before. However, dspy.configure(experimental=True) is likely to lead to much more predictable and accurate zero-shot generations with Chat LMs. Tested against several variants of GPT-3.5, GPT-4, Llama3, Mixtral, and DBRX.

All sorts of PRs over the past two weeks.

Most importantly, initial support for new adapters, with far more smooth support for Chat LMs.

The default behavior is unchanged from before. However, dspy.configure(experimental=True) is likely to lead to much more predictable and accurate zero-shot generations with Chat LMs. Tested against several variants of GPT-3.5, GPT-4, Llama3, Mixtral, and DBRX.

MIPROv2 and everything else from May 2024 and up to June 20th.