What changed in transformers, release by release
The lines under each release are transformers's own release-note bullets with pull request numbers, names and links removed, grouped by its own headings. Wording is theirs. Not affiliated with transformers.
v5.23.0
- Breaking๐จ Remove the
ep_routerexpert-parallel style - Breaking๐จ Run MXFP4 GPT-OSS experts through
ep_dispatch_experts - ChangedAdd inkling2
- ChangedAdd Jev + decision core
Patch release v5.22.2
- Fixed[Distributed] Set
device_idonly for NCCL - FixedHarden our Jinja sandboxing
Patch release v5.22.1
- ChangedWe just had to fix the documentation! [Fix] Remove WIP doc for
v5.22.0
- Breaking๐จ Separate EP logic into its own file
- New~1T total / ~52B active parameters: granular Mixture-of-Experts with 128 routed experts (4 active per token) and 1 shared expert.
- New1M context length, for long-context workloads.
- FixedForward token kwarg to
get_config_dictin PreTrainedConfig.from_pretrained - FixedPreserve NumPy integer label dtypes in default data collators
Patch release v5.21.1
- ChangedFix
rope_scalingsetter for configs that assign it during init We broke remote code loading in vllm, this fixes it sorry
v5.21
- BreakingRotating_light: Fix gemma4 cross imports resulting in import failing without torch but with PIL
- ChangedAdd Flint model support.
- ChangedAdd UltraVBERT
Patch release v5.20.2
- FixedFix inputs dict casting in generate
- FixedFix zero expert gradients in bfloat16 with expert parallelism and
ep_size<fsdp_size
v5.20.1
- FixedFix HunYuan-VL image assignment during batched generation
- FixedFix DeepGEMM Mega MoE under expert parallelism
- ChangedFix small bug in auto mappings
v5.20.0
- Breaking๐จ Remove deprecated cache utils methods
- ChangedAdd Muse Spark Open
- Changed[Model] Add native Ling 3.0 Flash support
v5.19.0
- Breaking๐จ Return router logits from every MoE model that computes them
- Breaking๐จ Select the OWLv2 image query by objectness
- Breaking๐จ Attention ๐จ Deprecate "paged|" prefix for SDPA and flash
- NewRevert "[CI] Temporarily disable AMD scheduled CI trigger"
- NewALM testx fIxing...
How we calculate
- Days between releases: The median of the gaps, in days, between the release dates listed above.
Data: huggingface/transformers's GitHub releases page, read November 18, 2026. Every other figure on this page is copied from the record, not calculated by us.
Get this for your repo, every release
transformers publishes release notes that people can read. FoxPlug writes yours from what you ship: a changelog entry and posts for the places your users are, each approved by you.
Get this for your repo, every releaseTry it on your repo nowYour changelog, written for you
FoxPlug writes your changelog from your releases and commits in plain English and publishes it as a page and a feed your users can follow. The page above shows what that looks like for a product people really use.
Write my changelog See the changelog FoxPlug wrote from its own commits
Free to start. Pro is a month.
FoxPlug Pro
- A public changelog page and feed for each project
- Written from your releases and commits, in plain English your users read
- Connect GitHub and new releases become new entries