Repository navigation
chore(deps): update loader dependencies non-major - #300
Merged
Merged
Conversation
dreadnode-renovate-bot
Bot
force-pushed
the
renovate/loader-deps
branch
from
October 14, 2026 00:47
b860e42 to
40ba6f4
Compare
dreadnode-renovate-bot
Bot
changed the title
| datasource | package | from | to | | ---------- | ------------ | ----------- | ----------- | | pypi | flash-attn | 2.8.3.post1 | 2.8.3.post2 | | pypi | pydantic | 2.13.5 | 2.14.1 | | pypi | transformers | 5.18.0 | 5.19.0 |
dreadnode-renovate-bot
Bot
force-pushed
the
renovate/loader-deps
branch
from
October 18, 2026 00:52
40ba6f4 to
4949d66
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR contains the following updates:
==2.8.3.post1→==2.8.3.post2==2.13.5→==2.14.1==5.18.0→==5.19.0Release Notes
Dao-AILab/flash-attention (flash-attn)
v2.8.3.post2Compare Source
Maintenance release of FlashAttention 2 on top of v2.8.3.post1. No kernel feature changes.
torch.load(..., weights_only=True)inflash_attn/utils/distributed.pyandflash_attn/models(CVE-2026-31253 / GHSA-7g5w-pq96-8c5w). #2974tarfileextraction inhopper/setup.py(CVE-2026-62239 / GHSA-w9r7-4gwr-j959). #2974dq_accumoffsets, which caused an illegal memory access or NaN gradients whenbatch * round128(seqlen_q) * nheads * headdimexceeded 2^31 (#2987). #2994setup.py: request cu13 prebuilt wheels for CUDA 13 torch.pydantic/pydantic (pydantic)
v2.14.1Compare Source
GitHub release
The code released in v2.5.0 is functionally identical to that of v2.5.0b1.
What's Changed
Packaging
pyright==1.1.335by @Viicos in #8075New Features
ValidationErrorerrors by @adriangb in #7806__doc__argument increate_model()by @chris-spann in #7863regex_engineflag - meaning you can use with the Rust or Python regex libraries in constraints by @utkini in #7768ComputedFieldInfoby @alexmojaki in #7889ruffformatter by @Luca-Blight in #7930validation_error_causeto config by @zakstucke in #7626CallableDiscriminatorandTagby @dmontagu in #7983CallableDiscriminatorrenamed toDiscriminatorby @dmontagu in #8047examplesandjson_schema_extrato@computed_fieldby @alexmojaki in #8013JsonValuetype by @dmontagu in #7998stras argument toDiscriminatorby @dmontagu in #8047SchemaSerializer.__reduce__method to enable pickle serialization by @edoakes in pydantic/pydantic-core#1006Changes
ultra_strictwith new smart union implementation, the way unions are validated has changed significantly to improve performance and correctness, we have worked hard to absolutely minimise the number of cases where behaviour has changed, see the PR for details - by @davidhewitt in pydantic/pydantic-core#867extra='allow'by @sydney-runkle in #7683Enumtypes with no cases by @sydney-runkle in #7927GenericbeforeBaseModelby @alexmojaki in #7891Performance
jiterby @samuelcolvin in pydantic/pydantic-core#974__getattr__for all package imports, improve import time by @samuelcolvin in #7947Fixes
mypyissue with subclasses ofRootModelby @sydney-runkle in #7677FieldInfowhen a forward ref gets evaluated by @dmontagu in #7698SecretStrfrom JSON (regression in v2.4) by @sydney-runkle in #7729defer_buildbehavior withTypeAdapterby @sydney-runkle in #7736mypyversions by @dmontagu in #7742TypeVarhandling when default is not set by @pmmmwh in #7719strictonEnumtype fields by @sydney-runkle in #7761weakref.refinstead of subclassing to fixcloudpickleserialization by @edoakes in #7780model_post_initin subclasses by @alexmojaki in #7775json_schema_extraby @alexmojaki in #7803strictspecification forUUIDtypes by @sydney-runkle in #7865pydantic.Field(kw_only=True)with inherited dataclasses by @PrettyWood in #7827validate_calldecorator for methods in classes with__slots__by @sydney-runkle in #7883dataclasses.fielddefault by @hramezani in #7898importlib_metadataon python 3.7 by @sydney-runkle in #7904|operator (Union) in PydanticRecursiveRef by @alexmojaki in #7892display_as_typeforTypeAliasTypein python 3.12 by @dmontagu in #7929NotRequiredgenerics inTypedDictby @sydney-runkle in #7932TypeAliasTypespecifications produce different schema definitions by @alexdrydew in #7893PrivateAttris passed fromAnnotateddefault position by @tabassco in #8004classmethodinstead ofclassmethod[Any, Any, Any]by @Mr-Pepe in #7979Optionalfield withvalidate_defaultonly performing one field validation by @sydney-runkle in pydantic/pydantic-core#1002definition-refbug withDictkeys by @sydney-runkle in pydantic/pydantic-core#1014booltypes withcoerce_numbers_to_str=Trueby @sydney-runkle in pydantic/pydantic-core#1017NaNin float and decimal constraints by @davidhewitt in pydantic/pydantic-core#1037lax_strandlax_intsupport for enum values not inherited from str/int by @michaelhly in pydantic/pydantic-core#1015UnionofListtypes by @sydney-runkle in pydantic/pydantic-core#1039max_digitsanddecimalsto pass if normalized or non-normalized input is valid by @sydney-runkle in pydantic/pydantic-core#1049ValidationErrormessages by @Iipin in pydantic/pydantic-core#1050'-'asdatetimeinput by @davidhewitt in pydantic/speedate#52 & pydantic/pydantic-core#1060New Contributors
pydanticpydantic-corev2.14.0Compare Source
GitHub release
What's Changed
The highlights of the v2.14 release are available in the blog post.
Several minor changes (considered non-breaking changes according to our versioning policy)
are also included in this release. Make sure to look into them before upgrading.
This release drops support for Python 3.9 and adds support for Python 3.15.
New Features
multiple_ofconstraint infractionandtimedeltacore schemas by @Viicos in #13873__namespace__parameter tocreate_model()by @Viicos in #13895Changes
ordered-dictcore schema by @Viicos in #13796countercore schema by @Viicos in #13824model_configattribute by @Viicos in #13825NameEmailvalidation by @Viicos in #13897Fixes
string_sub_typeunused error code by @Viicos in #13821OverflowErrorduringByteSizevalidation by @Viicos in #13850multiple_ofconstraints to be positive by @Viicos in #13862TypeErrorgracefully during docstring extraction by @Viicos in #13874TypedDicts in JSON Schema by @Viicos in #13891ser_json_timedeltato all datetime types in serialization inference by @Viicos in #13892initandkw_only=FalsefromField()to stdlib dataclass fields by @Viicos in #13898ZoneInfovalidation by @Viicos in #13912Decimalvalidation by @Viicos in #13940Fractionvalidation by @Viicos in #13944New Contributors
huggingface/transformers (transformers)
v5.19.0Compare Source
Release v5.19.0
New Model additions
EmbeddingGemma2
EmbeddingGemma 2 is a multimodal embedding model from Google built on the Gemma 4 architecture. It encodes text, images, audio, and video, individually or combined in one input, into a shared 768-dimensional vector space for cross-modal retrieval, semantic similarity, clustering, and classification. It uses Matryoshka Representation Learning, so embeddings can be truncated to 512, 256, or 128 dimensions. It also offers configurable visual and video token budgets, and unused vision or audio towers can be disabled at load time to save memory.
Links: Documentation
Breaking changes
All MoE models whose routers compute logits now return them when
output_router_logits=True, following the Qwen3-MoE pattern (arouter_logitsrecorder on the base model,MoeModelOutputWithPastfrom the backbone, and a MoE causal LM output from the head), so code that relied on the previous outputs or their absence should read the router logits from these output classes.Owlv2ForObjectDetection.embed_image_querynow selects the query box with the highest objectness score, as in the original OWLv2 notebook, instead of the OWL-ViT heuristic, so image-guided query embeddings and detections may differ from earlier releases.The
"paged|"prefix for SDPA and flash attention implementations is deprecated, so users should set the regular attention implementation (e.g.sdpaorflash_attention_2) for continuous batching instead ofpaged|sdpaorpaged|flash_attention_2.The regular flash and SDPA attention functions (
flash_attention.py,sdpa_attention.py) now support continuous batching directly, and"paged|..."implementations for these are redirected to them, while eager still requires the"paged|eager"prefix.In continuous batching, the cache update for the index-based and block-table paths is now fused into a single call, which slightly changes the cache update function's behavior and affects any custom code that calls the separate update paths.
Continuous batching internals changed in preparation for removing
"paged": `maxParallelization
Expert parallelism gains a token-dispatch implementation, selected via the new
ep_dispatch_expertsplan rule and now the default for Qwen3 MoE and Mellum, which removes the requirement that EP size equal TP size. TheTrainerwas also adapted to work with expert parallelism, and the docs now note that PEFT adapters support tensor parallelism. A CI-related fix for pipeline-parallel chart2table inference was also included.distributed]: adapt Trainer to work Expert Parallel (#48873) by @3outeille in [#48873]distributed] Add expert-parallel token dispatch, default for Qwen3 MoE (#48865) by @3outeille in [#48865]Cache
This release fixes quantized cache handling:
generateno longer mutates the user'scache_config, andQuantizedLayer.reorder_cacheis repaired. It also adds per-layer cache configuration, soDynamicCacheandStaticCacheinitialize each layer from its own config (sliding window, attention chunk size, conv states, and attention head counts) to better support heterogeneous models. Separately, the deprecation cycle on mask and cacheBugfixes and improvements
distributed] Default MoEep_plansto token dispatch (#49160) by @3outeille in [#49160].python-version(#49311) by @tarekziade in [#49311]distributed] decoupled tp_plan and ep_plan (#48859) by @3outeille in [#48859]distributed] Add dense and expert device mesh (#48857) by @3outeille in [#48857]labelsdocstring of CLIPSegForImageSegmentation (#49001) by @qgallouedec in [#49001]Mixin(#49105) by @zucchini-nlp in [#49105]torch_tpubackend (#49207) by @tengomucho in [#49207]GTE] Add memory mixin (#49234) by @vasqu in [#49234]Significant community contributions
The following contributors have made significant changes to the library over the last release:
GTE] Add memory mixin (#49234)Configuration
📅 Schedule: (UTC)
🚦 Automerge: Enabled.
♻ Rebasing: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox.
👻 Immortal: This PR will be recreated if closed unmerged. Get config help if that's undesired.
This PR has been generated by Mend Renovate CLI.