[dependabot]

Bumps the python-runtime group with 14 updates in the /backend directory:

Package	From	To
alembic	1.20.0	1.20.1
fastapi	0.143.0	0.162.0
google-cloud-storage	3.17.0	3.17.1
langchain-google-genai	4.4.1	4.4.2
langchain-google-vertexai	3.2.4	3.2.5
langgraph	1.2.14	1.2.17
langgraph-checkpoint	4.2.0	4.3.1
markdown	3.11	3.11.2
openai	3.29.0	3.37.0
sentence-transformers	6.1.0	6.2.0
transformers	5.19.0	5.20.2
sphinx	9.1.0	9.2.0
sqlalchemy	2.1.4	2.1.6
starlette	1.7.0	1.10.0
starlette	1.7.0	1.10.0

Bumps the python-runtime group with 3 updates in the /frontend directory: starlette, google-auth and pymongo.

Updates alembic from 1.20.0 to 1.20.1

Release notes

Sourced from alembic's releases.

1.20.1

Released: October 21, 2026

bug

• [bug] [autogenerate] Fixed issue where a table or column comment set to an empty string would cause autogenerate to detect a comment change on every run. As most backends, including PostgreSQL, MySQL, MariaDB and Oracle, don't store an empty string comment distinctly from no comment, autogenerate now treats an empty string comment the same as None when comparing comments. Additionally, an empty string comment is now rendered in op.create_table() and op.add_column() directives, where it was previously omitted. Pull request courtesy vineoy.

  References: #1085

• [bug] [autogenerate] [postgresql] Fixed rendering of the PostgreSQL ~sqlalchemy.dialects.postgresql.DOMAIN type in autogenerated migration scripts. Previously the type was rendered using repr(), which omitted the check, not_null, default, collation, constraint_name, schema and create_type arguments and did not apply a dialect prefix to the underlying data type. The full set of arguments is now rendered and the underlying data type is rendered using the standard type-rendering logic. Pull request courtesy Sebastian Cao.

  References: #1360

• [bug] [autogenerate] [postgresql] Fixed false-positive "Computed default cannot be modified" warning emitted during autogenerate when comparing PostgreSQL Computed columns whose reflected SQL expression includes type cast specifiers such as ::regconfig or ::text, which PostgreSQL adds to the expression as stored. These cast specifiers are now stripped before comparison so that otherwise equivalent expressions no longer trigger the warning. Pull request courtesy Sebastian Cao.

  References: #1462

• [bug] [autogenerate] [postgresql] Fixed issue where autogenerate would detect an unchanged PostgreSQL expression-based index as modified when the expression included type casts, such as a cast to a type with a length or precision, a nested cast, or a cast followed by a DESC modifier.

• [bug] [postgresql] Fixed bug where Operations.alter_column() against a PostgreSQL identity column would render an incomplete, invalid ALTER TABLE ... ALTER COLUMN statement with no clause at all when existing_server_default was omitted, even if there were no existing server default. Pull request courtesy Riti Grover.

... (truncated)

Commits

• See full diff in compare view

Updates fastapi from 0.143.0 to 0.162.0

Release notes

Sourced from fastapi's releases.

0.162.0

Omitted bool and bool | None form fields now use their declared defaults or produce missing-field errors, consistent with other fields and Pydantic behavior. FastAPI no longer infers False for omitted booleans. This applies to individual form parameters, form models, and unions of form models.

Omitted boolean model fields are excluded from model_dump(exclude_unset=True), and their defaults no longer count as submitted fields during union selection. Default factories and validate_default follow the usual rules.

For HTML checkboxes, declare a boolean with a default of False. See Boolean Form Fields.

Breaking Changes

• 🐛 Restore default handling for omitted boolean form fields. PR #16605 by @​tiangolo.

0.161.1

Fixes

• 🐛 Fix OpenAPI schemas for standalone nullable form unions. PR #16598 by @​tiangolo.

0.161.0

Features

• ✨ Add support for combining unions of form models. PR #16596 by @​tiangolo.

0.160.0

Query, header, cookie, and form parameter models now apply their extra configuration to all submitted fields in the same parameter location, consistent with Pydantic behavior. This includes fields declared in other models, individual parameters, or dependencies, which were previously excluded.

With extra="forbid", fields outside the model are rejected even if another parameter declares them. With extra="allow", those fields are included as model extras. With the default extra="ignore", they are ignored.

Breaking Changes

• 🐛 Apply each parameter model's extra configuration to sibling fields. PR #16594 by @​tiangolo.

Internal

• ⬆ Bump certifi from 2026.1.4 to 2026.10.19. PR #16571 by @​dependabot[bot].

0.159.1

Features

• ✨ Add support for optional unions of form models. PR #16586 by @​tiangolo.

0.159.0

Features

• ✨ Add support for optional query, header, and cookie parameter models. PR #16584 by @​tiangolo.

0.158.2

Features

• ✨ Add support for optional form models. PR #16582 by @​tiangolo.

0.158.1

... (truncated)

Commits

• 7a0a3f0 🔖 Release version 0.162.0 (#16606)
• b56f7c6 📝 Update release notes
• f02ef49 🐛 Restore default handling for omitted boolean form fields (#16605)
• 69ddc2b 🔖 Release version 0.161.1 (#16599)
• 0cafa07 📝 Update release notes
• f90d036 🐛 Fix OpenAPI schemas for standalone nullable form unions (#16598)
• 80c8742 🔖 Release version 0.161.0 (#16597)
• b105209 📝 Update release notes
• e42ac0a ✨ Add support for combining unions of form models (#16596)
• addd077 🔖 Release version 0.160.0 (#16595)
• Additional commits viewable in compare view

Updates google-cloud-storage from 3.17.0 to 3.17.1

Release notes

Sourced from google-cloud-storage's releases.

google-cloud-storage: v3.17.1

3.17.1 (2026-10-15)

Bug Fixes

• storage: set GRPC_ALTS_MAX_CONCURRENT_HANDSHAKES=2 in benchmark runner (#18703) (a9ca134)

Changelog

Sourced from google-cloud-storage's changelog.

Changelog

PyPI History

Commits

• 2d29c31 chore: release main (#18608)
• 7032a89 fix(gapic-generator): resolve showcase-mypy incompatible redefinition and add...
• 9025061 fix(bigquery): remove PendingDeprecationWarning from RowIterator.to_arrow (#1...
• 3743ff0 perf(spanner): avoid allocating empty RequestOptions on read and query paths ...
• a9ca134 fix(storage): set GRPC_ALTS_MAX_CONCURRENT_HANDSHAKES=2 in benchmark runner (...
• 09a3823 refactor(bigquery): move Arrow query result reading to _arrow_query (#18644)
• 939bdd0 feat(bigquery-storage): add Arrow options and cancel to read_rows (#18643)
• 8e640e0 chore(deps): update dependency nox to v2026 (#18681)
• 706ced0 chore(deps): update dependency shapely to v2.2.0 (#18694)
• af32b51 chore(deps): update dependency typing-extensions to v4.16.0 (#18697)
• Additional commits viewable in compare view

Updates langchain-google-genai from 4.4.1 to 4.4.2

Release notes

Sourced from langchain-google-genai's releases.

langchain-google-genai==4.4.2

Changes since libs/genai/v4.4.1

release(genai): 4.2.2 (#2058) fix(genai,vertexai): normalize model names and fix Gemini 3.5 cached content paths and streaming usage (#2044) chore(model-profiles): refresh model profile data (#1984)

Commits

• 69361c5 release(genai): 4.2.2 (#2058)
• ee823bf fix(genai,vertexai): normalize model names and fix Gemini 3.5 cached content ...
• f24b735 chore(model-profiles): refresh model profile data (#1984)
• 63eba68 chore: bump langgraph-sdk from 0.4.2 to 0.4.4 in /libs/community (#2052)
• 3d3b22f chore: bump langgraph-sdk from 0.4.2 to 0.4.4 in /libs/vertexai (#2054)
• 4e206c2 chore: bump fsspec from 2025.12.0 to 2026.6.0 in /libs/vertexai (#2051)
• See full diff in compare view

Updates langchain-google-vertexai from 3.2.4 to 3.2.5

Release notes

Sourced from langchain-google-vertexai's releases.

langchain-google-vertexai==3.2.5

Changes since libs/vertexai/v3.2.4

release(vertexai): 3.2.5 (#2080) fix(vertexai): drop unused pyarrow dependency (#2079) chore: remove precautionary third-party dependency caps (#2074) fix(vertexai): set ChatAnthropicVertex.profile from its Anthropic profiles (#2072) fix(vertexai): preserve citations from metadata-only streaming chunks (#2061) fix(vertexai): convert file content blocks to Anthropic documents (#2057) release(genai): 4.2.2 (#2058) fix(genai,vertexai): normalize model names and fix Gemini 3.5 cached content paths and streaming usage (#2044) chore(model-profiles): refresh model profile data (#1984) chore: bump langgraph-sdk from 0.4.2 to 0.4.4 in /libs/vertexai (#2054) chore: bump fsspec from 2025.12.0 to 2026.6.0 in /libs/vertexai (#2051) chore: bump urllib3 from 2.7.0 to 2.8.0 in /libs/vertexai (#2031) chore: bump anyio from 4.12.0 to 4.14.2 in /libs/vertexai (#2019) fix(vertexai): bump pyarrow dependency (#1962) chore(model-profiles): refresh model profile data (#1956) chore(model-profiles): refresh model profile data (#1906) chore: bump cryptography from 48.0.1 to 50.0.0 in /libs/vertexai (#1924) fix(vertexai): use Vertex profiles (#1913) chore: bump httplib2 from 0.31.0 to 0.32.0 in /libs/vertexai (#1908) chore: bump pyasn1 from 0.6.3 to 0.6.4 in /libs/vertexai (#1903) chore(model-profiles): refresh model profile data (#1879) chore: patch vertexai dependency alert (#1894) chore(model-profiles): refresh model profile data (#1802) chore: bump langgraph-checkpoint from 4.0.0 to 4.1.1 in /libs/vertexai (#1872) chore: bump langgraph-sdk from 0.3.9 to 0.3.15 in /libs/vertexai (#1873) chore: bump langsmith from 0.8.0 to 0.8.18 in /libs/vertexai (#1858) chore: bump vcrpy from 8.1.1 to 8.2.1 in /libs/vertexai (#1861) fix(vertexai): route labels to request envelope instead of GenerationConfig (#1865) fix(vertexai): replace removed bison integration test models (#1863) test(genai,vertexai,community): refresh Gemini model references (#1856) chore: bump cryptography from 46.0.7 to 48.0.1 in /libs/vertexai (#1850) test(vertexai): use content blocks in integration tests (#1847) test(vertexai): accept dict-form text content in gemini image output test (#1844) feat(genai,vertexai): add package version metadata (#1841) chore(genai,vertexai): bump min version of langchain-core (#1835)

Commits

• e013a09 release(vertexai): 3.2.5 (#2080)
• 79f9141 fix(vertexai): drop unused pyarrow dependency (#2079)
• 5576a98 chore: remove precautionary third-party dependency caps (#2074)
• fec2836 fix(vertexai): set ChatAnthropicVertex.profile from its Anthropic profiles ...
• f5f901a chore: bump aiohttp from 3.14.3 to 3.14.4 in /libs/genai (#2067)
• 800414b chore: bump aiohttp from 3.14.3 to 3.14.4 in /libs/community (#2066)
• da62c70 chore: bump langgraph-sdk from 0.4.4 to 0.4.6 in /libs/community (#2056)
• 880ceec chore: bump multidict from 6.7.0 to 6.9.1 in /libs/community (#2055)
• 3f1e01f chore: bump multidict from 6.7.1 to 6.9.1 in /libs/genai (#2053)
• 9f72e9a fix(vertexai): preserve citations from metadata-only streaming chunks (#2061)
• Additional commits viewable in compare view

Updates langgraph from 1.2.14 to 1.2.17

Release notes

Sourced from langgraph's releases.

langgraph==1.2.17

Changes since 1.2.16

• release(langgraph): 1.2.17 (#9418)
• fix(langgraph): apply an explicit recursion_limit equal to the default (#9417)
• fix(langgraph): keep the caller's namespace in RemoteGraph.stream(subgraphs=True) (#9401)
• chore(deps): remove precautionary third-party upper bounds (#9390)
• chore(deps): bump certifi from 2026.1.4 to 2026.10.19 in /libs/langgraph (#9375)
• chore(deps-dev): bump the minor-and-patch group in /libs/prebuilt with 6 updates (#9150)

langgraph==1.2.16

Changes since 1.2.15

• release(langgraph): 1.2.16 (#9334)
• fix(langgraph): let other tasks finish when a resume value is invalid (#9265)
• fix(langgraph): apply a resumed Command's update once (#9322)
• fix(langgraph): stop a subgraph from rerunning its finished tasks on resume (#9315)
• fix(langgraph): save a resumed exit run's pending writes once (#9314)
• release(prebuilt): 1.1.1 (#9316)
• release(checkpoint-postgres): 3.2.0 (#9313)
• feat(checkpoint-postgres): add prune to PostgresSaver and AsyncPostgresSaver (#9290)
• fix(langgraph): keep a DeltaChannel's value when its field changes to a plain reducer (#9302)
• release(checkpoint-postgres): 3.1.3 (#9287)
• release(checkpoint-sqlite): 3.1.2 (#9288)

langgraph==1.2.15

Changes since 1.2.14

• release(langgraph): 1.2.15 (#9283)
• fix(langgraph): give every message written to a messages DeltaChannel an id (#9275)
• fix(langgraph): save a cached node's writes like a node that ran (#9259)
• fix(langgraph): stop at an interrupt_before breakpoint when a drain is requested before it (#9281)
• fix(langgraph): keep an exit run from a checkpoint the thread moved past out of its newer branch (#9269)
• fix(langgraph): store an update_state input's writes on the checkpoint it updates (#9258)
• fix(langgraph): read a run's DeltaChannel input back only on its own checkpoints (#9260)
• fix(langgraph): snapshot a DeltaChannel that an update_state Overwrite resets (#9263)
• fix(langgraph): keep a fork's DeltaChannel snapshot from starting nodes that never ran (#9264)
• fix(langgraph): don't save a checkpoint past a failed DeltaChannel write (#9235)
• fix(langgraph,prebuilt): stop ToolNode from swallowing invalid resume values (#9232)
• fix(langgraph): never store exit-mode delta writes under the null task id (#9229)
• fix(langgraph): stop astream_events dropping control/interrupts on v1/v2 (#9219)
• chore(deps): bump notebook from 7.5.7 to 7.6.3 in /libs/langgraph (#9206)
• fix(langgraph): give each bulk_update_state update its own task id (#9128)
• fix(langgraph): replay a resumed exit-mode run's delta writes in live order (#9114)
• fix: order delta channel replay by task path (#8544)

Commits

• 4c52e12 release(langgraph): 1.2.17 (#9418)
• 6c4ceb1 fix(langgraph): apply an explicit recursion_limit equal to the default (#9417)
• 4897c51 fix(langgraph): keep the caller's namespace in RemoteGraph.stream(subgraphs=T...
• eb066bd release(cli): 0.4.38 (#9411)
• 6b83b9b release(sdk-py): 0.4.9 (#9398)
• 9d68e5b feat(cli): add the priority section to langgraph.json (#9393)
• 544386f feat(sdk-py): add Priority for thread priority settings (#9392)
• c2df8ab chore(deps): remove precautionary third-party upper bounds (#9390)
• 3cd57c0 fix(checkpoint): reject NUL bytes in namespace labels (#9383)
• 4098bb1 chore(deps): bump uv from 0.11.15 to 0.13.4 in /libs/cli (#9366)
• Additional commits viewable in compare view

Updates langgraph-checkpoint from 4.2.0 to 4.3.1

Release notes

Sourced from langgraph-checkpoint's releases.

langgraph-checkpoint==4.3.1

Changes since checkpoint==4.3.0

• release(checkpoint): 4.3.1 (#9447)
• fix(checkpoint): reject NUL bytes in namespace labels (#9383)
• chore(deps): bump certifi from 2026.1.4 to 2026.10.19 in /libs/checkpoint (#9372)

langgraph-checkpoint==4.3.0

Changes since checkpoint==4.2.0

• fix(checkpoint): default InMemorySaver delta history to the latest checkpoint (#9257)
• test(checkpoint): run the delta conformance suite against InMemorySaver in CI (#9261)
• fix(checkpoint,checkpoint-postgres,checkpoint-sqlite): validate namespace labels before store dispatch (#9217)
• fix(checkpoint): keep serialized data when an object can't be rebuilt (#9251)
• fix(langgraph): replay a resumed exit-mode run's delta writes in live order (#9114)
• fix: order delta channel replay by task path (#8544)
• chore(deps): bump the minor-and-patch group in /libs/checkpoint with 4 updates (#9148)
• chore(deps): bump urllib3 from 2.7.0 to 2.8.0 in /libs/checkpoint (#9131)
• chore(deps): bump anyio from 4.12.1 to 4.14.2 in /libs/checkpoint (#8995)
• chore(deps): bump the minor-and-patch group in /libs/checkpoint with 4 updates (#8777)
• fix(checkpoint): widen Store put value type to Mapping[str, Any] (#8617)

Commits

• 3521f08 release(checkpoint): 4.3.1 (#9447)
• 24294fa release(langgraph): 1.2.18 (#9445)
• f6317d5 release(sdk-py): 0.4.10 (#9446)
• 7e56e6c fix(langgraph): let error_handler recover on every path and stream failed tas...
• 0fc4f35 feat(langgraph)!: default delta snapshots to 100 updates (#9420)
• 58be691 release(cli): 0.4.40 (#9443)
• 1adeafe feat(cli): send the upload headers the control plane returns (#9396)
• dc1741c release(cli): 0.4.39 (#9442)
• c696d4e feat(cli): add --json to deploy list (#9395)
• 34c7f58 fix(sdk-py): retry runs.wait with a bodyless GET and raise run errors in sync...
• Additional commits viewable in compare view

Updates markdown from 3.11 to 3.11.2

Release notes

Sourced from markdown's releases.

Release 3.11.2

Fixed

• Sanitize attribute names in the fenced_code and legacy_attrs extensions so an attribute name cannot break out of its tag and inject markup (#1663).
• For backwards compatibility, revert the enforcement of CommonMark style punctuation rules (#1657).

Release 3.11.1

Fixed

• Update serializer to be non-recursive (#1644).
• Improve ancestor handling in the inline Treeprocessor (#1646).
• Keep a raw HTML comment inside an inline HTML element that closes in the same paragraph, instead of splitting the paragraph around it (#1643).
• Fix issue where inline HTML attributes were rejected if they had < or > in the attribute (#1647).
• Fix issue where an unterminated end tag (</foo) could cause all remaining content to be dropped (#1651).
• Complete rework of emphasis handling to improve nested emphasis handling better (#1632).

Changelog

Sourced from markdown's changelog.

[3.11.2] - 2026-10-19

Fixed

• Sanitize attribute names in the fenced_code and legacy_attrs extensions so an attribute name cannot break out of its tag and inject markup (#1663).
• For backwards compatibility, revert the enforcement of CommonMark style punctuation rules (#1657).

[3.11.1] - 2026-10-13

Fixed

• Update serializer to be non-recursive (#1644).
• Improve ancestor handling in the inline Treeprocessor (#1646).
• Keep a raw HTML comment inside an inline HTML element that closes in the same paragraph, instead of splitting the paragraph around it (#1643).
• Fix issue where inline HTML attributes were rejected if they had < or > in the attribute (#1647).
• Fix issue where an unterminated end tag (</foo) could cause all remaining content to be dropped (#1651).
• Complete rework of emphasis handling to improve nested emphasis handling better (#1632).

[3.11.0] - 2026-09-25

Changed

• Inline processors now resume searching after the previous match, improving performance for repeated inline patterns (#1619).
• Officially support Python 3.15 and drop support for Python 3.10
• Walk backtick runs in BacktickInlineProcessor without a regex (#1620).
• Switch static site generator for documentation from MkDocs to Zensical (#1627, #1635, #1637, and #1638).

Fixed

• Ensure removing Abbreviations does not raise an error (#1634).
• Fix an issue with excessive backtracking when matching inline code blocks (#1617).
• md_in_html now honors tags added to Markdown.block_level_elements after the extension is loaded (#1246).
• Fix quadratic-time regex backtracking in ReferenceProcessor when a link reference definition has no URL, e.g. a line consisting only of [id]: followed by many trailing spaces (#798).
• Document attr_list usage for def_list (#1123).

[3.10.3] - 2026-07-30

Fixed

• Fix SetextHeaderProcessor regex to prevent mixed = and - chars in setext-style headers (#1606).
• Add AI Policy to Contributing Guide.
• Officially document all included extensions as being in maintenance mode.

... (truncated)

Commits

• b41e0aa Bump version to 3.11.2
• 934c53e Add checklist of parts to PR template
• 2249ffd Update AI Policy
• a27428b Sanitize attribute names in fenced_code and legacy_attrs
• 90cdcf5 Add fixed category to changelog
• 97233ef Revert enforcement of CommonMark punctuation rules
• 2258b48 Bump version to 3.11.1
• 6f825aa Rewrite emphasis handling
• 0bf535b Keep a comment inside raw inline HTML in the same paragraph
• 3ade80d Add playground to docs
• Additional commits viewable in compare view

Updates openai from 3.29.0 to 3.37.0

Release notes

Sourced from openai's releases.

v3.37.0

3.37.0 (2026-10-22)

⚠ BREAKING CHANGES

• api: This minor release includes a breaking change to Decisions types. In Live response models, question instructions can now be a string or a list of structured text, image, or JSON parts; check isinstance(question.instructions, str) before using string operations and handle the structured form. Decisions request input types also include additional message roles and tool-history items, so update exhaustive type checks accordingly. Existing string instructions and DecisionInputMessageParam user-message inputs remain supported.

Features

• api: expand Decisions input and question types (#4158) (1e45ad1)

Chores

• api: preserve file and video polling helpers (#4155) (7db9d88)
• api: preserve upload helpers and Live connection defaults (#4152) (a29dc39)
• deps: bump astral-sh/setup-uv from 10.2.0 to 10.3.0 (#4109) (fa9a161)

v3.36.1

3.36.1 (2026-10-22)

Bug Fixes

• api: honor per-call WebSocket query settings (#4138) (d40a5e1)
• examples: stop microphone capture between recordings (#4143) (26e5d92)
• validate browser calls before reading driver state (#4142) (2bfe460)

Chores

• api: clarify service-account API key deletion (#4151) (6ed945e)

Refactors

• prepare handwritten chunked upload runtime (#4146) (2d58e82)
• prepare handwritten vector-store polling runtime (#4148) (7cd7028)

v3.36.0

3.36.0 (2026-10-21)

Features

• add Vertex AI support for the Responses API (#4139) (3e63d33)
• api: add 30-minute cache-write usage counts (#4111) (5589146)
• api: add compaction trigger IDs (#4122) (0e5794d)

... (truncated)

Changelog

Sourced from openai's changelog.

3.37.0 (2026-10-22)

⚠ BREAKING CHANGES

• api: This minor release includes a breaking change to Decisions types. In Live response models, question instructions can now be a string or a list of structured text, image, or JSON parts; check isinstance(question.instructions, str) before using string operations and handle the structured form. Decisions request input types also include additional message roles and tool-history items, so update exhaustive type checks accordingly. Existing string instructions and DecisionInputMessageParam user-message inputs remain supported.

Features

• api: expand Decisions input and question types (#4158) (1e45ad1)

Chores

• api: preserve file and video polling helpers (#4155) (7db9d88)
• api: preserve upload helpers and Live connection defaults (#4152) (a29dc39)
• deps: bump astral-sh/setup-uv from 10.2.0 to 10.3.0 (#4109) (fa9a161)

3.36.1 (2026-10-22)

Bug Fixes

• api: honor per-call WebSocket query settings (#4138) (d40a5e1)
• examples: stop microphone capture between recordings (#4143) (26e5d92)
• validate browser calls before reading driver state (#4142) (2bfe460)

Chores

• api: clarify service-account API key deletion (#4151) (6ed945e)

Refactors

• prepare handwritten chunked upload runtime (#4146) (2d58e82)
• prepare handwritten vector-store polling runtime (#4148) (7cd7028)

3.36.0 (2026-10-21)

Features

• add Vertex AI support for the Responses API (#4139) (3e63d33)
• api: add 30-minute cache-write usage counts (#4111) (5589146)
• api: add compaction trigger IDs (#4122) (0e5794d)
• api: add custom voice consent management (#4135) (de6d70d)
• types: prepare native synchronous WebSocket option type (#4118) (90c0bac)

... (truncated)

Commits

• e8a0fd2 release: 3.37.0 (#4154)
• fa9a161 chore(deps): bump astral-sh/setup-uv from 10.2.0 to 10.3.0 (#4109)
• 1e45ad1 feat(api): expand Decisions input and question types (#4158)
• 1f76a59 ci: allow time for browser dependency installation (#4159)
• 7db9d88 chore(api): preserve file and video polling helpers (#4155)
• a29dc39 chore(api): preserve upload helpers and Live connection defaults (#4152)
• 49e6757 release: 3.36.1 (#4145)
• 6ed945e chore(api): clarify service-account API key deletion (#4151)
• 7cd7028 refactor: prepare handwritten vector-store polling runtime (#4148)
• d40a5e1 fix(api): honor per-call WebSocket query settings (#4138)
• Additional commits viewable in compare view

Updates sentence-transformers from 6.1.0 to 6.2.0

Release notes

Sourced from sentence-transformers's releases.

v6.2.0 - Model merging, faster loading, and distributed evaluation

This release adds model merging for all four model types, reduces redundant Hub requests when loading models, and distributes evaluator inference across DDP ranks. It also includes lower-memory training losses, more flexible multi-vector compression, and fixes throughout inference, training, and evaluation.

Install this version with

# Inference only, use one of:
pip install sentence-transformers==6.2.0
pip install sentence-transformers[openvino]==6.2.0
Multimodal dependencies (optional):
pip install sentence-transformers[image]==6.2.0
pip install sentence-transformers[audio]==6.2.0
pip install sentence-transformers[video]==6.2.0
Training + Inference
pip install sentence-transformers[train]==6.2.0
pip install sentence-transformers[merge]==6.2.0
Or combine as needed:
pip install sentence-transformers[train,image]==6.2.0

Model merging for all four model types (#3776)

Combine compatible fine-tuned checkpoints without another training run using SentenceTransformer.merge, SparseEncoder.merge, CrossEncoder.merge, or MultiVectorEncoder.merge. The integration uses Arcee AI's mergekit and supports well over a dozen methods such as Linear, SLERP, TIES, and DARE.

Install the merge extra with pip install "sentence-transformers[merge]". For example, take a weighted average of two compatible MiniLM models:

from sentence_transformers import SentenceTransformer
models = [
"sentence-transformers/all-MiniLM-L6-v2",
"sentence-transformers/multi-qa-MiniLM-L6-cos-v1",
]
merged = SentenceTransformer.merge(
models,
method="linear",
parameters={"weight": [0.6, 0.4]},
)
merged.save_pretrained("merged-MiniLM-L6")
embeddings = merged.encode(["Combining compatible model weights."])

Or interpolate between the same inputs with SLERP:

</tr></table> 

... (truncated)

Commits

• 0b6937d Release v6.2.0
• 45682ed Update documentation announcement for v6.2 (#4243)
• aed573b Forward gradient checkpointing options to Transformers (#4241)