Hugging Face Release Notes

New Models

Mistral Large 4 (Le Chonk)

Mistral Large 4 is Mistral's largest and most capable model to date: a 1 trillion-parameter natively multimodal Mixture-of-Experts model with 49 billion active parameters per token (52 billion including embeddings and output layers), trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own datacenters in Europe.

Mistral Large 4 unifies instruction following, reasoning, coding, and agentic workflows in a single set of weights. While a generalist model, it shows especially good performance on workloads such as cybersecurity, finance, and law, and it surpasses frontier closed models on visual grounding. It is natively fluent in 160+ languages, including every official language of the European Union.

Key Features

Mistral Large 4 includes the following architectural choices:

• ~1T total / ~52B active parameters: granular Mixture-of-Experts with 128 routed experts (4 active per token) and 1 shared expert.
• 1M context length, for long-context workloads.
• Multimodal input: accepts text and image input, with text output. Pixtral-style vision encoder trained to handle variable image sizes and aspect ratios.
• Instruct and Reasoning functionalities with function calls, reasoning effort configurable per request.

Mistral Large 4 offers the following capabilities:

• Reasoning Mode: three reasoning efforts — low, medium, high (default) — so the same model can answer a quick chat reply or work through a complex agentic run.
• Vision: analyzes images, charts, technical drawings, and documents, with state-of-the-art visual grounding among open weights.
• Multilingual: natively fluent in 160+ languages, including all official languages of the European Union.
• System Prompt: strong adherence and support for system prompts.
• Agentic: native function calling, JSON output, and strict JSON schema enforcement.
• Cybersecurity: state of the art among open-weight models for security work, while refusing malicious requests.
• Large Context Window: supports a 1M context window.

The model is released under the Mistral Ecosystem Open-Weight License: a modified version of the MIT license, for both commercial and non-commercial use — operating a Model-as-a-Service business requires a separate commercial license with Mistral.

Recommended Settings

Reasoning Effort:

'low' / 'medium' / 'high' — default is 'high'.

Use reasoning_effort="high" for complex prompts and agentic usage, 'low' for fast chat replies and 'medium' for day to day usage.

Sampling: for all reasoning efforts, we recommend the following sampling parameters:

• temperature=1.0
• top_p=0.95

System Prompt: define a clear environment and use case, including guidance on how to effectively leverage tools in agentic systems.

Tools: keep the set of tools well-defined and limit their number to the minimum required for the use case.

Context: the checkpoint supports a 1M context window.

Breaking changes

• 🚨 Separate EP logic into its own file (#49679) by @3outeille

Quantization

FP8 quantized models can now be used for adapter training (such as LoRA with PEFT). Backward passes were added to the dense and grouped FP8 matmuls, and is_trainable is now True. Backward also dequantizes experts in chunks and skips autograd at inference, which cuts peak memory in the MoE layer backward from about 28.9 GiB to 2.8 GiB at 8k tokens. Separately, the AMD CI image now install

• FP8 backward: chunked expert dequant, skip autograd.Function at inference (#50117) by @ArthurZucker in [#50117]
• Train adapters on top of FP8 weights (#50031) by @qgallouedec in [#50031]
• [AMD CI] Install onnxscript, bitsandbytes and peft in the AMD image (#50046) by @Abdennacer-Badaoui in [#50046]

Bugfixes and improvements

• Processor unnecessary warnings (#49812) by @zucchini-nlp in [#49812]
• Fix boolean padding in ClapFeatureExtractor (#23648) (#49960) by @rudy-07 in [#49960]
• [Fix] Count cropping of an uninitialized layer as a no-op (#50104) by @remi-or in [#50104]
• [loading] Allow PermuteForRope to act on single tensors while renaming (#50092) by @Cyrilvallez in [#50092]
• Add Nemotron 3.5 Transcribe (#50084) by @eustlb in [#50084]
• [Nemotron3Diarization] add left context support (#49283) by @eustlb in [#49283]
• [Janus] Fix processor iterating a string prompt character by character (#50044) by @tarekziade in [#50044]
• fix(config): forward token kwarg to get_config_dict in PreTrainedConfig.from_pretrained (#50081) by @ydshieh in [#50081]
• Fix rope_scaling setter for configs that assign it during init (#50080) by @hmellor in [#50080]
• [timesfm] Fix fp16 overflow on pad_val in _forward_transform (#50021) by @tarekziade in [#50021]
• Update dev version (#50068) by @albertvillanova in [#50068]
• Pin the shared Hub MIC fleet dataset (#49802) by @tarekziade in [#49802]
• [docs] NVIDIA/Intel/CPU (#49650) by @stevhliu in [#49650]
• Document how to handle non-persistent buffer initialization in custom models (#49880) by @Tialo in [#49880]
• Remove legacy benchmark/ and disable automatic benchmark runs (#50040) by @tarekziade in [#50040]
• Remove deprecation cycle for candidate generators (#50034) by @Cyrilvallez in [#50034]
• Fix test pil import leak (#49915) by @guarin in [#49915]
• fix: preserve NumPy integer label dtypes in default data collators (#49900) by @BlazeSR10hawk in [#49900]
• [ESMFold2] Don't build the MSA encoder for checkpoints without one (#49890) by @jimburtoft in [#49890]
• [minimax_m3_vl] Gate the integration tests on total accelerator memory (#50025) by @tarekziade in [#50025]
• Fix: Add kwargs in LFM2.5 conv layers (#50022) by @younesbelkada in [#50022]
• Forward code_revision to every pipeline loader (#49832) by @Rocketknight1 in [#49832]
• [moshi] Skip Moshiko integration tests whose checkpoint is gone (#50017) by @tarekziade in [#50017]
• Keep a capturing group's escapes when reversing a conversion pattern (#49772) by @IlyasMoutawwakil in [#49772]
• perf(ci): skip the lazy blob fetch in git blame on a partial clone (#50007) by @nrenvo in [#50007]
• fix(canary): set layerdrop=0.0 in test tester to prevent flaky test (#50012) by @ydshieh in [#50012]
• add a WA to Dockerfile for XPU (#50008) by @kaixuanliu in [#50008]
• pipelines: defer eager model imports in text_to_audio and token_classification (#50001) by @ydshieh in [#50001]
• [exporters] resolve @register_patch paths lazily to avoid eager model imports (#50000) by @ydshieh in [#50000]
• [Fix] Remove useless kwargs (#49966) by @remi-or in [#49966]
• Fix model's config attr naming (#49898) by @zucchini-nlp in [#49898]

Significant community contributions

The following contributors have made significant changes to the library over the last release:

• @tarekziade
  • Fix _init_weights gaps in 7 audio models and enable test_can_init_all_missing_weights for them (#49386)
  • Fix image backend equivalence tests for Pillow 13 (#50105)
  • [Janus] Fix processor iterating a string prompt character by character (#50044)
  • [timesfm] Fix fp16 overflow on pad_val in _forward_transform (#50021)
  • Pin the shared Hub MIC fleet dataset (#49802)
  • Remove legacy benchmark/ and disable automatic benchmark runs (#50040)
  • [minimax_m3_vl] Gate the integration tests on total accelerator memory (#50025)
  • [moshi] Skip Moshiko integration tests whose checkpoint is gone (#50017)
• @eustlb
  • Add Nemotron 3.5 Transcribe (#50084)
  • [Nemotron3Diarization] add left context support (#49283)

This is a security release. See GHSA-4prv-5hj7-c2xr for details.

This patch release fixes how a component's declared library name is resolved when loading with from_pretrained.

Pipelines and models declare each component as a ["<library>", "<class>"] pair in model_index.json / config.json. This release changes how that library name is resolved across the DiffusionPipeline, AutoModel, and ModularPipeline loading paths, and adds a clearer error when the declared class is not found in the resolved module.

All users on 0.42.0 are encouraged to upgrade:

pip install diffusers==0.42.1

Release v5.21.0
New Model additions
MiniCPM-OCR
MiniCPM-OCR is a document-understanding model from OpenBMB's MiniCPM-V family. It keeps the MiniCPM-V 4.6 architecture, which pairs a SigLIP vision encoder and a window-attention merger with a Qwen3.5 language model. It is specialized for reading text-dense images such as scanned pages, screenshots, receipts and tables.
Links: Documentation | Paper
Feat test (#49920) by @guarin in #49920
Flint
Flint is Recursive's mixture-of-experts language model. Compared to a standard SwiGLU MoE transformer such as Qwen3-MoE, it uses hybrid attention (sliding-window attention with rotary embeddings over the current token and the previous 512 tokens in most layers, and full attention without positional encoding in the remaining layers). It also has learned attention sinks, as in GPT-OSS, with per-head RMS-normalized queries and keys. Each sparse layer routes a token to 6 of 256 experts by sigmoid score and adds an ungated shared expert, and the first two layers are dense.
Links: Documentation
Add Flint model support. (#49824) by @lovishmadaan in #49824
UltraVBert
UltraVBert extends ModernVBert by pairing the UltraBERT text encoder with a SigLIP vision encoder for visual document understanding and retrieval. Its image processor splits each page into 512-pixel tiles plus a global view. The connector merges each 4 x 4 block of vision patches into one token, so a tile becomes 64 image tokens, and RMS-normalizes them to the scale of UltraBERT's token embeddings. The max_image_tiles setting (27 by default) caps the number of tiles per page.
Links: Documentation
Add UltraVBERT (#49685) by @athrael-soju in #49685
Gradium Phonon TTS
Gradium Phonon TTS is Gradium's family of on-device text-to-speech models, trained with the Pocket TTS recipe. A causal transformer reads a voice prompt and the text, then emits one continuous 32-dimensional audio latent per step at 12.5 Hz through a one-step flow head, and an end-of-speech head decides when the utterance ends. The latents are decoded to 24 kHz audio by a Pocket TTS audio tokenizer. Unlike Pocket TTS, voice conditioning uses 512-dimensional latents from a separate, unreleased speaker codec, so voices come as precomputed latents and cannot be cloned from a recording.
Links: Documentation
Add GradiumPhononTTS (#49791) by @eustlb in #49791
Pocket TTS
Pocket TTS is Kyutai's 100M-parameter text-to-speech model, built as a Continuous Audio Language Model that runs faster than real time on CPU. A causal transformer emits one continuous audio latent per frame (12.5 Hz) through a one-step flow head, and an end-of-speech head decides when to stop. A Mimi-based audio tokenizer encodes voice prompts and decodes the generated latents to 24 kHz audio. Two flow heads are released, Lagrangian self-distillation (default) and drifting, and voices can be cloned from a recording, passed as precomputed latents, or loaded from cached KV states.
Links: Documentation | [Paper](https://huggingface.co/papers/2509.06926 (the Continuous Audio Language Models paper, arXiv 2509.06926). It is the only paper the documentation links, and it describes the architecture family Pocket TTS belongs to. The docs do not cite a separate Pocket TTS technical report.) | [Blog Post](NONE. The documentation links no official blog post. It points only to the original code repository at https://github.com/kyutai-labs/pocket-tts.)
add PocketTTS (#49589) by @eustlb in #49589
Pocket TTS Audio Tokenizer
The Pocket TTS audio tokenizer is the codec used by Pocket TTS. It has Mimi's architecture (SEANet convolutions plus a small causal transformer on each side), but its residual vector quantizer is replaced by a continuous 32-dimensional bottleneck, the VAE-GAN from Continuous Audio Language Models. It converts 24 kHz audio into latents at 12.5 Hz and decodes them back to audio. It serves as the audio_tokenizer of PocketTTSProcessor, encoding voice prompts and decoding generated latents, and its encoder and decoder can also be loaded on their own.
Links: Documentation | Paper
add PocketTTSAudioTokenizer (#49588) by @eustlb in #49588
Breaking changes
Fixed Gemma4 processor imports (the Gemma4, Gemma4 unified and EmbeddingGemma2 processors) so from transformers import * no longer fails in environments that have PIL but not torch; no user action is needed.
🚨 Fix gemma4 cross imports resulting in import failing without torch but with PIL (#49748) by @Sai-Suraj-27
Added an "fp8" backend to QuantizedCache, alongside quanto and hqq, that stores KV states in torch.float8_e4m3fn with a single scale per tensor; enable it with cache_implementation="quantized" and cache_config={"backend": "fp8"}.
🚨Add an FP8 backend to QuantizedCache (#48677) by @jiqing-feng
Tensor parallelism now has a dedicated embedding style, shards every model's input embeddings by default, and supports widths that aren't divisible by the TP size, so ColwiseParallel and RowwiseParallel now apply only to linear layers; custom TP plans that applied them to embeddings must switch to the new embedding style.
🚨 TP: give embeddings their own style, shard input embeddings by default, and shard widths that don't divide by the TP size (#49570) by @IlyasMoutawwakil
Every MoE TP and EP plan now shards its experts, so models no longer keep all experts on every rank, and a new test_moe_parallel_plans_shard_experts test enforces this; custom or downstream MoE parallel plans must add expert sharding to pass it and to avoid wasted memory.
🚨 MoE: shard the experts in every TP/EP plan, and test for it (#49206) by @IlyasMoutawwakil
Cache
This release improves cache handling across the library. Flash Attention now works with static caches, BLOOM no longer crashes with static cache or compiled generate, chunked prefill supports cache continuation, and the Qwen4-Exp indexer's position history stays aligned with its cached keys. Linear-attention caches also gain new getters (is_recording_past, get_recurrent_state, conv-state access) that stop exposing cache internals and prepare for paged-c
[FA] Fix FA with static cache (#49998) by @Cyrilvallez in [#49998]
[docs] CB cache behavior (#49025) by @stevhliu in [#49025]
[CB] Reduce warmup time by not computing attention over full cache (#49853) by @remi-or in [#49853]
Fix Qwen4-Exp indexer position cache lifecycle (#49840) by @taking-lying-flat in [#49840]
[Cache] Linear attn cache: use new getters (#49375) by @remi-or in [#49375]
[Cache] Get conv states (#49309) by @remi-or in [#49309]
[Cache] Add get_recurrent_state (#49306) by @remi-or in [#49306]
[Cache] Add is_recording_past (#49305) by @remi-or in [#49305]
Fix BLOOM crash with static cache / compiled generate (#49481) by @kaixuanliu in [#49481]
Allow Cache continuation for chunked prefill (#49662) by @Cyrilvallez in [#49662]
Vision
Vision image processors now composite PNG transparency onto white instead of exposing black pixels, and background_color is preserved correctly across the DeepSeek-VL and Janus processors. Model fixes cover SmolVLM's inputs_merger when image tokens aren't a multiple of the patch size, HunYuan-VL image assignment during batched beam search, and Ernie4.5-VL-MoE expert sharding under TP/EP. The DETR
Composite transparency in shared RGB conversion and image processors (#49079) by @Nikhi00718 in [#49079]
[Fetcher] Fix missing test deps for dynamically-resolved image processors (#49826) by @ydshieh in [#49826]
fix(vision): Route the DETR conversion scripts through from_pretrained (#49830) by @harshaljanjani in [#49830]
Fix background_color handling across DeepSeek-VL and Janus image processors (#49767) by @tritsystem in [#49767]
Move the MI300 docker image to ROCm 10.1 and update test expectations (#49460) by @Abdennacer-Badaoui in [#49460]
docs: fix 'allows to' grammar in Vision Encoder Decoder docs (#48730) by @mrchatam in [#48730]
Fix SSH runner: add /opt/venv/bin to PATH for Python 3.11 images (#49759) by @ydshieh in [#49759]
Fix SmolVLM inputs_merger when image tokens are not a multiple of patch_size (#49725) by @albertvillanova in [#49725]
Ernie4.5-VL-MoE: shard the text and vision experts under TP and EP (#49462) by @IlyasMoutawwakil in [#49462]
Fix HunYuan-VL image assignment during batched generation (#49557) by @Arnav66692 in [#49557]
Generation
This release fixes a range of generation bugs: BLOOM crashes with static cache or compiled generate, generate dropping the type of dict-like inputs such as ModelOutput, and incorrect image assignment or placeholder handling in batched or beam-search generation for HunYuan-VL and Idefics2/3. It also fixes Qwen3-Omni audio generation mode handling, SAM mask-generation multi-zoom crops, empty frame s
fix: prevent empty frame sampling for short videos (#49908) by @BlazeSR10hawk in [#49908]
Fix Windows path separators breaking relative imports of Hub custom-generate modules (#48951) by @dltsum in [#48951]
Fix multi-zoom crops in the SAM mask-generation pipeline and related bugs (Fixes #35530) (#49689) by @stlukyanenko in [#49689]
update speculative decoding docs (#49697) by @merveenoyan in [#49697]
fix idefics3 generation issues (#49738) by @molbap in [#49738]
fix tests for generation (#49740) by @molbap in [#49740]
Fix inputs dict casting in generate (#49731) by @Cyrilvallez in [#49731]
[fix][Qwen3-Omni] Respect generation_mode in audio generation (#49505) by @Sunt-ing in [#49505]
Parallelization
This release extends parallelism support and fixes several related bugs. It adds a tensor-parallel plan for Nemotron and makes Ernie4.5-VL-MoE shard its text and vision experts under TP and EP. It also restores expert-parallel token-dispatch gradients on torch 2.7–2.10, fixes a Molmo2 model-parallel device mismatch, and corrects base_model_prefix in several model families (PP-OCR detectors
Add tensor parallel plan for Nemotron (#48780) by @Siddhant231xyz in [#48780]
fix model parallel bug for molmo2 model (#49816) by @kaixuanliu in [#49816]
Fix base_model_prefix in the PP-OCR detectors, CLIPSeg, TimesFM, CLIP and VideoPrism heads so base checkpoints load (#48871) by @blipbyte in [#48871]
Fix expert-parallel token dispatch gradients on torch 2.7 to 2.10 (#49710) by @digit50 in [#49710]
Skip Muse TP (#49717) by @remi-or in [#49717]
Quantization
Quantization now handles DeepSeek-V4.1's FP4 experts correctly: FineGrainedFP8Config gains an expert_dtype option ("fp8" by default, or "fp4") that the FP8 experts read instead of allocating them as FP8. TimesFM 2.5 no longer casts float inputs to an integer weight dtype when loaded with quantization such as bitsandbytes 4/8-bit. A broader cleanup
Fine-grained FP8: DeepSeek-V4's expert_dtype in the quantization config (#49848) by @IlyasMoutawwakil in [#49848]
Fix dtype cast in TimesFM 2.5 ResidualBlock for quantized weights (#46942) by @JOhnsonKC201 in [#46942]
Quantization cleanpup (#49548) by @ArthurZucker in [#49548]
Bugfixes and improvements
[Fix] Fix CSM codebook embedding tying (#49197) by @orrangetabby17 in [#49197]
[tests_fetcher] fix missing .as_posix() in preprocessor_patterns path keys (#49994) by @ydshieh in [#49994]
fix(ci): skip bot suggestion if a maintainer already posted run-slow: (#49992) by @ydshieh in [#49992]
[loading] Allow Renaming and Converter to work on the sume subkeys (#49928) by @Cyrilvallez in [#49928]
[Fix] Fix NUMA and cpuset detection in CB (#49980) by @remi-or in [#49980]
[Refactor] Make CB utils depend on no CB import (#49968) by @remi-or in [#49968]
chore: improve typing in modeling_rope_utils.py (#49831) by @RyanMullins in [#49831]
is_fouroversix_available() returned an (available, version) tuple. Ev… (#49894) by @sywangyi in [#49894]
Match a skip-list entry from the start only up to a module boundary (#49771) by @IlyasMoutawwakil in [#49771]
[pocket_tts, gradium_phonon_tts] use sub_configs_defaults (#49868) by @eustlb in [#49868]
Fixes a dtype mismatch in flash_attn test (#49912) by @kaixuanliu in [#49912]
[Fix] Remove duplicate assignment in DeepseekV3.2 config (#49896) by @remi-or in [#49896]
add XPU Expectations for 4 new model tests (#49815) by @kaixuanliu in [#49815]
Make sub-config-defaults easier (#49906) by @zucchini-nlp in [#49906]
Fix out-of-bounds index bug in musicgen_melody model (#49842) by @kaixuanliu in [#49842]
test: add parity check to the CPU-only continuous batching test (#49728) by @nrenvo in [#49728]
[CB] Reduce the max_memory_percent in case of a CPU model (#49844) by @remi-or in [#49844]
Improve clarity and grammar in Auto Classes documentation (#44550) by @himani27301 in [#44550]
[gradium_phonon_tts] point to the released Hub repos (#49867) by @eustlb in [#49867]
[EmbeddingGemma2] update integration tests (#49576) by @eustlb in [#49576]
Fix a comment (#44872) by @IvanFan-Van in [#44872]
Fix hardcoded torch.float64 crashing on MPS (#46723) (#46730) by @Abhijeet1815 in [#46730]
fixup metadata workflow (#49854) by @vasqu in [#49854]
docs(whisper): cast input features to model dtype in AutoModel example (#49837) by @Kayvan-Zahiri in [#49837]
Fix torch_float casting to int instead of float (#47468) by @rahulrshetty45 in [#47468]
Upgrade Python Min version to 3.11 (#49251) by @Sai-Suraj-27 in [#49251]
skip sdpa_can_dispatch_on_flash tests for 3 models (#49809) by @kaixuanliu in [#49809]
[LED][Longformer] Replace for-loop with unfold in _chunk ONNX-export path (#46169) by @guinik in [#46169]
logits_to_keep bool mask for all ForXXX classes (#48825) by @jp1924 in [#48825]
Fix several object detection models/loader fix for module tying (#49827) by @molbap in [#49827]
Fix encoder logits processors when precomputed encoder_outputs lack token IDs (#49696) by @kunalsanga in [#49696]
Defer some imports (#49663) by @jayzuccarelli in [#49663]
refactor CTC loss (#49833) by @ebezzam in [#49833]
Add Intel XPU CI (#49254) by @regisss in [#49254]
fix(mask2former): register input_projections as nn.ModuleList (fixes #49698) (#49703) by @Omarinho12345 in [#49703]
Only tie MM Grounding DINO decoder heads when the config shares them (#49552) by @ in [#49552]
[Docs] Add Marathi (मराठी) translation of README (#48746) by @shreeyadav06 in [#48746]
Fix CLVP c_proj residual scaled-init silently never applied (#47478) by @ShamSaleem in [#47478]
fix failed test cases for unlimited_ocr model (#49806) by @kaixuanliu in [#49806]
Fix transformers env when requests is not installed (#48412) by @panshularora in [#48412]
Apply EXIF orientation in load_image_as_tensor with torchvision (#49794) by @digit50 in [#49794]
Fix Fuyu processor test (#49821) by @zucchini-nlp in [#49821]
use torchcodec-xpu and upgrade torch to 2.14.1 (#49814) by @sywangyi in [#49814]
update some case to add xpu expectation (#49807) by @sywangyi in [#49807]
[serge] Fix 2 integration tests regressed by commit c48824e (PR #49404) (#49705) by @sergereview[bot] in [#49705]
docs: fix OOM/hang for Qwen dynamic resolution models by setting min_pixels and max_pixels (#46007) by @poojansatani in [#46007]
docs: fix 'allows to' grammar in DPT docs (#48728) by @mrchatam in [#48728]
docs: fix 'allows to' grammar in LiLT docs (#48731) by @mrchatam in [#48731]
docs: fix prompt formatting and code alignment in Gemma model doc (#48775) by @Rattan344 in [#48775]
docs: fix grammar typo in ALBERT model card (#48774) by @Rattan344 in [#48774]
Deprecate old input names (#49424) by @zucchini-nlp in [#49424]
Fix NameError in CI comment workflows: missing section marker variables (#49801) by @ydshieh in [#49801]
[CI] Improve run-slow PR comments and remove CI recap (#49782) by @ydshieh in [#49782]
Show the actual accelerate version for transformers env (#49668) by @Sai-Suraj-27 in [#49668]
[CI] fix perceriver old regression (#49792) by @molbap in [#49792]
mimi moshi output capturing (#49670) by @eustlb in [#49670]
Remove the resolved prototype Whisper MIC expected failure (#49785) by @tarekziade in [#49785]
Fix CohereCompass text RoPE to use Qwen3-VL interleaved M-RoPE (#49715) by @mmangkad in [#49715]
Alias thinking -> reasoning_content (#49535) by @Rocketknight1 in [#49535]
[vLLM] Add Qwen-Omni Thinker base models and Gemma 3n per-layer inputs (#49437) by @harshaljanjani in [#49437]
Update MIC fleet references and documentation (#49768) by @tarekziade in [#49768]
[CI] no split modules for SAM (#49766) by @molbap in [#49766]
Enable tests forgotten in idefics3 (#49758) by @molbap in [#49758]
Fix tests_fetcher backbone step (never worked since introduction) (#49742) by @ydshieh in [#49742]
[modular] Extend the "del self.xxx" pattern to match typed assignments (#49735) by @remi-or in [#49735]
Xpu add (#49757) by @sywangyi in [#49757]
Deprecate rope_scaling assignment and normalize legacy RoPE keys (#49515) by @jinwukong in [#49515]
Make longrope compileable (#49555) by @jiqing-feng in [#49555]
Fix stack_frames > 1 wiring bug in MiniCPMV4_6VideoProcessor (#49752) by @airlsyn in [#49752]
fix moshi failing tests (#49750) by @eustlb in [#49750]
[docs] Gaudi (#49618) by @stevhliu in [#49618]
[serge] Fix 2 integration tests for model convnextv2 failing with output_mismatch (tensor values differ (2)) (#48838) by @sergereview[bot] in [#48838]
[serge] Fix 2 integration tests for model eomt_dinov3 failing with output_mismatch (tensor values differ (2)) (#48882) by @sergereview[bot] in [#48882]
[serge] Fix 4 integration tests for model lw_detr failing with output_mismatch (tensor values differ (4)) (#49040) by @sergereview[bot] in [#49040]
[serge] Fix 2 integration tests for model nemotron failing with output_mismatch (list output differs (2)) (#48945) by @sergereview[bot] in [#48945]
multi device fix for molmo2 (#49736) by @molbap in [#49736]
Fix the rest of VLMs (#49720) by @zucchini-nlp in [#49720]
Fix ValueError when loading a fusion-enabled checkpoint twice (#49412) by @kaixuanliu in [#49412]
Avoid a host-to-device copy in eager_mask (#49684) by @hmellor in [#49684]
is_tracing() now recognizes XPU stream capture, alongside the exist… (#49722) by @sywangyi in [#49722]
Document how to avoid the Metaspace prefix after added tokens (#45822) by @qflen in [#45822]
Declare the input modalities of the VLMs that report text only (#49677) by @qgallouedec in [#49677]
Fix zero expert gradients in bfloat16 with expert parallelism and ep_size < fsdp_size (#49688) by @qgallouedec in [#49688]
Fix test selection for virtual module imports (#49690) by @tarekziade in [#49690]
[docs] add Google TPU hardware guide (#49642) by @tengomucho in [#49642]
Copy default response parser args (#49370) by @Rocketknight1 in [#49370]
Make DtensorShardOperation work on torch 2.6 and 2.7 (#49657) by @hazelxue in [#49657]
Fix from_pretrained on torch 2.7: require torch>=2.8 for distributed checkpointing (#49652) by @naveecode in [#49652]
Fix DeepGEMM Mega MoE under expert parallelism (#49428) by @IlyasMoutawwakil in [#49428]
Fix flaky Evolla tests producing NaN outputs (#49660) by @tarekziade in [#49660]
[rocm] Add ROCm expectations to Kolibri-1 logits test (#49661) by @ericschreiber in [#49661]
[CB] Various small fixes for MPS (#49654) by @remi-or in [#49654]
Aligns test expectations with the accelerator path under test: (#49653) by @sywangyi in [#49653]
[docs] Processor config (#49643) by @stevhliu in [#49643]
post release stuff (#49646) by @vasqu in [#49646]
Significant community contributions
The following contributors have made significant changes to the library over the last release:
@ydshieh
[tests_fetcher] fix missing .as_posix() in preprocessor_patterns path keys (#49994)
fix(ci): skip bot suggestion if a maintainer already posted run-slow: (#49992)
[Fetcher] Fix missing test deps for dynamically-resolved image processors (#49826)
Fix NameError in CI comment workflows: missing section marker variables (#49801)
[CI] Improve run-slow PR comments and remove CI recap (#49782)
Fix tests_fetcher backbone step (never worked since introduction) (#49742)
Fix SSH runner: add /opt/venv/bin to PATH for Python 3.11 images (#49759)
@remi-or
[Fix] Fix NUMA and cpuset detection in CB (#49980)
[Refactor] Make CB utils depend on no CB import (#49968)
[Fix] Remove duplicate assignment in DeepseekV3.2 config (#49896)
[CB] Reduce warmup time by not computing attention over full cache (#49853)
[CB] Reduce the max_memory_percent in case of a CPU model (#49844)
[modular] Extend the "del self.xxx" pattern to match typed assignments (#49735)
[Cache] Linear attn cache: use new getters (#49375)
[Cache] Get conv states (#49309)
[Cache] Add get_recurrent_state (#49306)
[Cache] Add is_recording_past (#49305)
[Kernels] Add MPS to supported device types (#49655)
Skip Muse TP (#49717)
[CB] Various small fixes for MPS (#49654)
@Nikhi00718
Composite transparency in shared RGB conversion and image processors (#49079)
@eustlb
[pocket_tts, gradium_phonon_tts] use sub_configs_defaults (#49868)
[gradium_phonon_tts] point to the released Hub repos (#49867)
[EmbeddingGemma2] update integration tests (#49576)
Add GradiumPhononTTS (#49791)
add PocketTTS (#49589)
add PocketTTSAudioTokenizer (#49588)
🟡 [mimi] separate encoder and decoder (#49577)
mimi moshi output capturing (#49670)
fix moshi failing tests (#49750)
@jp1924
logits_to_keep bool mask for all ForXXX classes (#48825)
@molbap
Fix several object detection models/loader fix for module tying (#49827)
[CI] fix perceriver old regression (#49792)
[CI] no split modules for SAM (#49766)
Enable tests forgotten in idefics3 (#49758)
fix idefics3 generation issues (#49738)
fix tests for generation (#49740)
multi device fix for molmo2 (#49736)
@athrael-soju
Add UltraVBERT (#49685)
@harshaljanjani
fix(vision): Route the DETR conversion scripts through from_pretrained (#49830)
fix(audio): Route the ALM conversion scripts through from_pretrained (#48525)
[vLLM] Add Qwen-Omni Thinker base models and Gemma 3n per-layer inputs (#49437)
@lovishmadaan
Add Flint model support. (#49824)
@tarekziade
Remove the resolved prototype Whisper MIC expected failure (#49785)
Update MIC fleet references and documentation (#49768)
Cleaned up tokenizer tests (#49669)
Fix test selection for virtual module imports (#49690)
Fix flaky Evolla tests producing NaN outputs (#49660)
@yonigozlan
Fix small bug in auto mappings (#49664)
@ArthurZucker
Quantization cleanpup (#49548)

Tip

This release brings FLUX.3 Image Klein, Bria Fibo 2.0, Kandinsky 6, SANA-WM, and RFD4-Proteína to Diffusers, alongside new adapter support, hardware integrations, and modular pipelines.

🚨 Python 3.11 or newer is now required.

FLUX.3 Image Klein

FLUX.3 Image Klein from Black Forest Labs supports text-to-image generation and editing with one or more reference images. Both the base and step-distilled checkpoints are available through ModularPipeline, with an integrated prompt upsampler that turns short requests into detailed captions and supports layout instructions with bounding boxes. The integration also includes single-file transformer loading and BFL FP8r and NVFP4 checkpoints.

Load the pipeline with ModularPipeline.from_pretrained(...), then load its weights with load_components(...). See the FLUX.3 documentation for examples and guidance on prompt upsampling, editing, and memory use.

Bria Fibo 2.0

Bria Fibo 2.0 supports image generation from structured JSON captions and editing with up to five reference images through BriaFibo2Pipeline. The integration includes the dev checkpoint and the edit-turbo checkpoint, which runs in four steps without classifier-free guidance. See the Fibo 2.0 documentation for the supported prompt formats.

LoRA loading and DreamBooth LoRA training examples are included for text-to-image and image editing. Thanks to @muhammadali-afridi for the model integration (#14991).

Kandinsky 6

Kandinsky 6 generates video with synchronized audio from text or a reference image through Kandinsky6TI2VAPipeline. A separate Kandinsky6SRPipeline provides video super-resolution. The integration supports Pro and Lite generation checkpoints, including distilled variants, and introduces PiflowScheduler for the distilled models. See the Kandinsky 6 documentation for checkpoint-specific settings and super-resolution requirements. Thanks to @Lefff for the model integration (#14949).

SANA-WM

SANA-WM generates video from a starting image, a text prompt, and a camera trajectory. Control the camera with explicit poses or a WASD/IJKL action string, then optionally refine the output with the separate LTX-2 refiner. See the SANA-WM documentation for the two-stage workflow.

Thanks to @lawrence-cj for the integration (#13881).

RFD4-Proteína

RFD4-Proteína from the Baker Lab and NVIDIA brings biomolecular design to Diffusers. It jointly generates protein sequences and all-atom structures while conditioning on molecular context such as binding targets, motifs, or ligands. The integration includes the pipeline, transformer, autoencoder, schedulers, and LoRA loading. Preparing model inputs and writing CIF files uses the separate rfd4-proteina-prep package; see the RFD4-Proteína documentation.

Thanks to @kierandidi for the integration (#15088).

Other highlights

• Qwen-Image 2.1: new modular pipelines and tensor parallelism support. (#14976, #14865)
• More hardware support: TorchTPU integration covers eager execution, torch.compile, and tensor parallelism. Tensor parallelism also gains Intel XPU support. (#14039, #15013)
• LoKr adapters: load LyCORIS Kronecker-product adapters for Z-Image and FLUX.2, including Klein. (#14163)
• LTX-2 motion control: motion_speed controls the positional time base independently of playback frame rate. Use a compatible adapter for speeds far from the default, such as the LTX-2.5 slow-motion control LoRA. (#14967)
• KVAE-Audio: AutoencoderKLKVAEAudio adds encoding and decoding for 48 kHz audio. (#14384)
• Caching: MagCache now supports FLUX.2. (#14756)
• Adapter fixes: Z-Image honors the LoRA scale in joint_attention_kwargs, Qwen-Image applies .alpha scaling to lora_A/lora_B checkpoints, and text encoders correctly report active adapters. (#15015, #15016, #14961)
• Pipeline fixes: preserve explicit timesteps in FlowMatchEulerDiscreteScheduler, fix Anima text encoder exports and standalone core denoising, and preserve the three-value return from InpaintProcessor.preprocess when no mask is supplied. (#14956, #14943, #14481)

Deprecations and compatibility

• Python 3.10 support is dropped. Upgrade to Python 3.11 or newer before installing this release. (#15030)
• Modular latent outputs: FLUX, Krea 2, and LTX-Video now unpack latents in their core denoise groups. Custom workflows should pass unpacked latents to their decode blocks. The fallback for packed latents is deprecated for removal in 0.47.0. (#14737)
• Hub compatibility: the supported huggingface_hub range is now >=1.32.0,<3.0. (#14863)
• TorchAO imports: having a TorchAO version older than 0.15 installed no longer breaks model imports. TorchAO quantization still requires version 0.15 or newer. (#15095)

Thanks to everyone who contributed to this release!

All commits

• fix #14461 by @yiyixuxu in #14956
• Kandinsky6 by @sayakpaul in #14949
• Bump doc-builder workflow pin by @paulinebm in #14957
• [docs] Jobs guide by @stevhliu in #14926
• [docs] Community methods by @stevhliu in #14973
• Fix active adapter detection for text encoders by @Subodh-17 in #14961
• fix(image_processor): maintain 3-tuple return contract in InpaintProcessor.preprocess when mask is None by @patrickswedish in #14481
• [ci] consolidate labeling on PRs. by @sayakpaul in #14911
• [LoRA] add LoKr adapter support (Z-Image, Flux2/Klein) by @christopher5106 in #14163
• [TPU] TorchTPU backend integration - eager / torch.compile / tp by @JingyaHuang in #14039
• [cli] detect composite block classes in custom_blocks by @yiyixuxu in #15001
• Fix Anima text encoder exports and standalone core denoising by @lucasruan1618 in #14943
• Bump huggingface_hub upper bound to <3.0 (diffusers) by @Wauplin in #14863
• Make Stable Diffusion device-map comparisons deterministic by @sywangyi in #15010
• xpu supports TP as well by @sywangyi in #15013
• Add SANA-WM camera-controlled image-to-video pipeline by @lawrence-cj in #13881
• [Z-Image] Apply the LoRA scale passed via joint_attention_kwargs by @Sergasgr in #15015
• [utils] preserve method metadata in apply_forward_hook by @javierdejesusda in #13803
• Update FlashAttention Hub integration and guard CUDA-only tests by @sywangyi in #15024
• [hooks] Support Flux2 in MagCache by @rootonchair in #14756
• 🔒 Pin GitHub Actions to commit SHAs by @paulinebm in #15028
• [CI] Clean up test_modeling_common.py by @DN6 in #14988
• ci: move to python 3.11. by @sayakpaul in #15030
• [core] Tensor parallelism for Qwen-Image-2.1 by @JingyaHuang in #14865
• refactor: drop the Neuron RoPE path in Qwen Image by @JingyaHuang in #15042
• add modular pipeline for qwen-image 2.1 by @akshan-main in #14976
• Flux3 image by @sayakpaul in #15046
• [docs] FLUX.3: prompt upsampler required, ours or any LLM by @stephen37 in #15059
• FLUX.3 prompt upsampler block by @yiyixuxu in #15058
• [docs] FLUX.3: modular-only notes, lazy loading, callbacks by @yiyixuxu in #15069
• FLUX.3: match the released BFL reference image sizing by @yiyixuxu in #15073
• post-release 0.41.0 by @sayakpaul in #15062
• move unpack out of the flux, krea2 and ltx decoders into the core denoise groups by @akshan-main in #14737
• [feat] add KVAE-Audio autoencoder by @ivnvalex in #14384
• Fix PyArrow compatibility in CPU CI by @sayakpaul in #15082
• Add Bria fibo-2 by @muhammadali-afridi in #14991
• Add RFD4-Proteina by @kierandidi in #15088
• [tests] xfail some kandinsky6 tests by @sayakpaul in #14977
• Fix model imports when torchao is older than 0.15 by @usmanmateen in #15095
• [ci] use python 3.11 when using metadata. by @sayakpaul in #15093
• [ci] fix misnamed test steps in push_tests.yml by @sayakpaul in #15094
• [LTX-2] Add motion_speed to condition the positional time base apart from the playback frame rate by @christopher5106 in #14967
• [ci] skip bf16 precision test on T4. by @sayakpaul in #15091
• Use HF Trusted Publisher token in update_metadata workflow by @coyotte508 in #15096
• [LoRA] apply .alpha to Qwen-Image lora_A/lora_B LoRAs by @dcaayushd in #15016
• [ci] fix compilation CI. by @sayakpaul in #15092
• add fibo-2 LoRA training and loading by @linoytsaban in #15084
• Release: v0.42.0-release by @sayakpaul (direct release commit)

Significant community contributions

The following contributors added models and capabilities in this release:

• @muhammadali-afridi: Bria Fibo 2.0 integration. (#14991)
• @lawrence-cj: SANA-WM camera-controlled video generation and refinement. (#13881)
• @kierandidi: RFD4-Proteína biomolecular design integration. (#15088)
• @ivnvalex: KVAE-Audio autoencoder. (#14384)
• @christopher5106: LoKr adapter support and LTX-2 motion-speed control. (#14163, #14967)
• @akshan-main: Qwen-Image 2.1 modular pipelines and consistent unpacked latent outputs for modular denoise groups. (#14976, #14737)
• @sywangyi: Intel XPU tensor parallelism, FlashAttention Hub integration updates, and deterministic device-map tests. (#15013, #15024, #15010)
• @rootonchair: MagCache support for FLUX.2. (#14756)

Full changelog: v0.41.0 to the v0.42.0 release commit

OpenUSD files (.usd, .usda, .usdc and .usdz) can now be viewed on the Hub. Click one in a dataset, model or bucket to open it in an interactive 3D viewer, which loads its layers, payloads and textures. For example, an LG laptop, a Vertiv server rack or a robot arm from BEHAVIOR-1K.

Datasets with SimReady packages also show a 3D Assets preview on their page, with the number of packages from across the dataset, as on LGElectronics/simready-assets and Vertiv/simready-assets.

The viewer is open source and available in huggingface.js as @huggingface/usd-web-viewer.

Patch release: v5.20.2

This patch notably solves an issue where some inputs could loose their original type inside prepare_inputs_for_generation in generate, leading to problems with multimodal models. It also fixes some other issues listed on the below commits.

It contains the following commits:

• Fix inputs dict casting in generate (#49731) by @Cyrilvallez
• Fix zero expert gradients in bfloat16 with expert parallelism and ep_size < fsdp_size (#49688) by @qgallouedec
• Declare the input modalities of the VLMs that report text only (#49677) by @qgallouedec
• Avoid a host-to-device copy in eager_mask (#49684) by @hmellor

Release v5.20.1

Special surprise including the new model Beam from Reflection AI 🥳

Beam is a token-efficient, frontier-scale model focused on coding and agentic use cases. It is built for teams that want to run a powerful model on their own hardware, infrastructure, or cloud, and customize it to their specific needs.

Links: Documentation

Fix small bug in auto mappings (#49664) by @yonigozlan in #49664

Some real patch fixes

We broke loading models on older torch versions 😢 should be fixed now. But here is the full list of fixes

• Fix HunYuan-VL image assignment during batched generation (#49557) by @Arnav66692
• Fix DeepGEMM Mega MoE under expert parallelism(#49428) by @IlyasMoutawwakil
• Fix from_pretrained on torch 2.7: require torch>=2.8 for distributed checkpointing (#49652) by @naveecode
• Make DtensorShardOperation work on torch 2.6 and 2.7 (#49657) by @hazelxue

Full Changelog: v5.20.0...v5.20.1

Release v5.20.0

New Model additions

Muse Spark Open

Muse Spark Open is a multimodal mixture-of-experts model from Meta that takes text, images, audio, and video as input and generates text. It has 816B total parameters (42B active per token) and a context window of up to 1,048,576 tokens shared across modalities. It is trained for agentic coding tasks such as multi-file editing, debugging, and tool use. Checkpoints are available in BF16 and NVFP4 formats.

Links: Documentation | Blog Post

Add Muse Spark Open (#49644) by @guarin in #49644

Ling 3.0

Ling 3.0 Flash is a hybrid Mixture-of-Experts language model from InclusionAI. Its decoder groups five Kimi Delta Attention (KDA) layers with one Multi-head Latent Attention (MLA) layer. The MLA blocks use interleaved rotary embeddings and a head-wise output gate, and the feed-forward blocks combine routed and shared experts. It is natively supported in Transformers as BailingHybridForCausalLM, so it loads without trust_remote_code=True.

Links: Documentation

[Model] Add native Ling 3.0 Flash support (#49446) by @icavan in #49446

Kolibri1

Kolibri 1 is Aleph Alpha's mixture-of-experts reasoning model, focused on German and English. Each of its 50 layers routes every token to 6 of 384 experts using sigmoid routing, and adds one shared expert. Four sliding-window attention layers alternate with one full-attention layer, and RoPE is applied only in the sliding-window layers.

Links: Documentation

Add Kolibri 1 (#49463) by @lucamotz in #49463

MiniCPM-V 4.7

MiniCPM-V is a series of efficient multimodal large language models developed by OpenBMB. Like MiniCPM-V 4.6, the MiniCPM-V 4.7 architecture pairs a SigLIP vision encoder that has a window-attention merger with a Qwen3.5 language model backbone, and supports both 4x and 16x visual downsampling modes.

The main addition over 4.6 is canvas M-RoPE: instead of numbering visual tokens along a single 1-D sequence, the model lays every image out on a 2-D canvas and assigns each visual token a (temporal, height, width) position, so slices of the same image keep their spatial relationship and video frames keep their temporal order.

Fix more stuff (#48979) by @zucchini-nlp in #48979

UnlimitedOcr

UnlimitedOcr is a single 3B-parameter OCR model from Baidu with a standard 32,768-token context length. It builds on DeepSeek OCR and replaces all decoder attention layers with Reference Sliding Window Attention (R-SWA). This keeps the KV cache constant during decoding, so the model can transcribe dozens of pages in a single forward pass. It supports "gundam" mode (640x640 tiles with dynamic cropping) and "base" mode (a single 1024x1024 view), and it can output region detections with bounding boxes.

Links: Documentation | Paper

Add Unlimited OCR (#46836) by @guarin in #46836

UltraBERT

UltraBERT is a family of bidirectional text encoders trained with a combination of masking ratios, pruning, and distillation for classification and retrieval tasks. The released encoders support sequences of up to 32,768 tokens. They use rotary position embeddings, RMSNorm, SwiGLU feed-forward layers, and alternating local and global attention (every third layer is global). Checkpoints include UltraBERT-150M and UltraBERT-380M for masked language modeling and fine-tuning, and UltraBERT-Embed-150M for text embeddings.

Links: Documentation | Paper

Integrate the UltraBERT architecture (#49596) by @tomaarsen in #49596

MOSS-Transcribe-Diarize

MOSS-Transcribe-Diarize 0.9B is an end-to-end audio understanding model for long-form multi-speaker transcription, speaker diarization, and timestamps. It combines a Whisper-style encoder, a 4× frame merge step, a multi-modal projector, and a Qwen3 language model. It produces joint transcription and diarization in a single pass, with segments formatted as [start][S01]text[end]. It supports long-form audio through Whisper-window chunking in the processor, promptable transcription with hotword biasing, and 50+ languages.

Links: Documentation

Add MOSS-Transcribe-Diarize model (#48386) by @itazap in #48386

Breaking changes

🚨 General breaking notice that inference with EP has regressions in speed; we are working on fixing this ASAP! 🚨

Deprecated checks and methods have been removed from cache_utils.py (affecting models such as minimax_m3_vl and pi0), so any code still calling those deprecated cache methods must switch to the current cache API.

🚨 Remove deprecated cache utils methods (#49605) by @guarin

Accidental imports from other models were removed from several model files (e.g. embedding_gemma2, ijepa, lightglue), and the modular converter now avoids adding such cross-model imports. If you imported these names through those model modules, import them from their original modules instead.

🚨 Remove imports from other models (#49473) by @guarin

The minimum kernels version is now 0.17.2, which allows use_fallback=KernelizeFallback.ALL so that a failed kernel load falls back to the original implementation instead of raising an error. Upgrade kernels to 0.17.2 or later.

🚨 Bump min kernels to 0.17.2 for better fallback (#49235) by @SunMarc

A new SubConfigSpec centralizes the initialization of sub-configs, so default sub-configs are collected from dataclass fields instead of by instantiating the class, and this touches the configuration utilities and many model configs. Custom or existing nested configs may need to be updated to declare their sub-configs through this mechanism.

🚨 SubConfigSpec for centralized init of subconfigs (#48282) by @zucchini-nlp

The LUKE and mLUKE tokenizers now treat entity special tokens such as <ent> as special tokens (not split, honored by skip_special_tokens, normalized) while keeping the same spacing as regular vocabulary tokens when decoding. Tokenization and decoding output involving these tokens may therefore differ from before, so check any code or saved outputs that depend on the old behavior.

🚨🚨🚨 luke fixe (#48408) by @itazap

Exporters

Transformers exporters now support quantization directly during export across Dynamo, ONNX, OpenVINO, and ExecuTorch. The new unified quantization API integrates PT2E, ONNX Runtime, and NNCF quantizers, enabling static and dynamic quantization as well as weight-only compression (INT8/INT4), with configurable calibration workflows. This makes it easier to export optimized models for efficient inference across different backends.

For more information see the docs: https://huggingface.co/docs/transformers/exporters_quantization

Parallelization

This release improves distributed training and inference. FSDP × TP model and optimizer checkpoints can now be saved and loaded across different parallel configurations, and expert-parallel dispatch, tensor-parallel plans, static-cache generate(), and the caching allocator's byte count were fixed. Tied LM heads in composite models now get their own TP plan entry, enable_expert_parallel applies the EP plan alone again, TPU loading uses the `tor

• FSDP x TP Saving / Loading, Optimizer (#49256) by @michaelbenayoun in [#49256]
• FSDP x TP Saving / Loading (#48802) by @michaelbenayoun in [#48802]
• Don't set a device per rank for gloo tensor-parallel tests (#49636) by @IlyasMoutawwakil in [#49636]
• fix(distributed): keep enable_expert_parallel as EP alone (#49633) by @dacorvo in [#49633]
• Fix a port allocation race in FSDP tests (#49455) by @icavan in [#49455]
• Expert-parallel dispatch hands the experts the received rows, not the all-to-all's pending result (#49421) by @IlyasMoutawwakil in [#49421]
• Fix static-cache generate() with host-side inputs and under TP (#49317) by @dacorvo in [#49317]
• Composite lm head tp plan (#49080) by @IlyasMoutawwakil in [#49080]
• Fix: use torch.accelerator device APIs so TP loading works again on TPU (#49502) by @tengomucho in [#49502]
• Count ep and fsdp in caching_allocator_warmup byte calculation (#49384) by @3outeille in [#49384]
• Do not wrap the model in FSDP for expert-parallel token dispatch (#49383) by @3outeille in [#49383]

Generation

This release fixes static-cache generate() bugs with host-side inputs and under tensor parallelism. It also speeds up Whisper generation on TPU by cutting device syncs and recompilation, and fixes Whisper batched timestamp decoding for tensor and NumPy inputs. Pipelines get a cleanup of generation-argument handling that removes spurious warnings.

• Add job to automatically generate Intel CPU Docker image (#49560) by @regisss in [#49560]
• Image decoding moves to torchcodec (#47963) by @zucchini-nlp in [#47963]
• Optimize Whisper generation for TPU and device sync (#49426) by @tengomucho in [#49426]
• Generation cleanup in pipelines (#49269) by @Rocketknight1 in [#49269]
• Fix Whisper batch timestamp decoding for tensors (#49419) by @Kingofbooks in [#49419]

Vision

Vision processing was updated in several places. Image decoding now uses torchcodec, Molmo2's image and video kwargs follow the library's standard conventions, and Pixtral Flash Attention no longer detects false image boundaries after the axial RoPE refactor, which had caused CUDA illegal memory access errors and incorrect output. Several CI and Docker changes also shipped: GPU images moved to Python 3.11, a numpy/pyarrow conflict

• Fix numpy/pyarrow conflict in the XPU Docker image (#49582) by @jiqing-feng in [#49582]
• CI: Move GPU images to python 3.11 (#49352) by @tarekziade in [#49352]
• Update molmo2 processing images/videos kwargs (#49563) by @guarin in [#49563]
• Fix Pixtral Flash Attention image boundaries after axial RoPE refactor (#49373) by @oliverholworthy in [#49373]

Kernels

This release improves kernel support across hardware. flash_attention_2 now falls back to the AITER kernel on ROCm when flash-attn isn't installed, and T5-style models get a contiguous SDPA position-bias mask so they use fused kernels again. The Hub rotary kernel is now limited to CUDA inference because its backward pass gave incorrect gradients. The Atlas Inference Gated DeltaNet kernels for Strix Halo and

• Fall back to the AITER kernel for flash_attention_2 on ROCm (#49608) by @Abdennacer-Badaoui in [#49608]
• Make the SDPA position_bias mask contiguous so T5-style models use fused kernels (#49517) by @IvenHsu01 in [#49517]
• Use the Hub rotary kernel for inference only on CUDA (#49439) by @qgallouedec in [#49439]
• [Atlas Inference] Enable torch.compile for the Strix Halo Gated DeltaNet hub kernel (#49453) by @AzeezIsh in [#49453]
• [Atlas Inference] Fix GB10 Gated DeltaNet hub kernel for transformers 5.17+ and enable torch.compile (#49452) by @AzeezIsh in [#49452]

Cache

This release fixes several cache-related bugs: static-cache generate() now handles host-side nested inputs and tensor parallelism correctly, the offloaded cache prefetch no longer races with the offloading copy, and Phi3 LongRoPE cache rebuilds at the context boundary now receive the full prefix. It also counts the query token in MiMo-V2-Flash's sliding window and corrects the quantized cache (hqq/quanto) documentation exam

• Fix offloaded cache prefetch racing with the offloading copy (#49363) by @qgallouedec in [#49363]
• docs: update quantized cache hqq/quanto examples and documentation (#47212) by @DavidJohnQuinlan in [#47212]
• [MiMo-V2-Flash] Count the query token in the sliding window (#49170) by @guapisolo in [#49170]
• Fix Phi3 LongRoPE cache rebuild at the context boundary (#49361) by @LE0-Lin in [#49361]

Bugfixes and improvements

• [docs] Hardware docs template (#49575) by @stevhliu in [#49575]
• Add MIC contract lock files for PR CI integration (#49631) by @ydshieh in [#49631]
• Use aiter-rope version 3 on ROCm (#49595) by @Abdennacer-Badaoui in [#49595]
• Remove missing methods from docs (#49637) by @guarin in [#49637]
• Translate torch.mm(..., out_dtype=...) to ONNX (#49635) by @IlyasMoutawwakil in [#49635]
• [docs] Hardware section (#49541) by @michaelbenayoun in [#49541]
• Raise when a torch fallback would drop packed-sequence boundaries (#49474) by @qgallouedec in [#49474]
• Fa packed sequence decode (#49610) by @SunMarc in [#49610]
• fix nested set_attn_implementation (#49138) by @eustlb in [#49138]
• feat: Add MTP support to Mellum2/2.1 (#49531) by @shadeMe in [#49531]
• extend onnx exporters tests to non-cuda devices (#49627) by @kaixuanliu in [#49627]
• Document video processor backends (#49470) by @DimensionSTP in [#49470]
• Make sinusoidal position embeddings compileable (#49482) by @jiqing-feng in [#49482]
• Save the PEFT adapter of a load-time sharded model from its own parameters (#49500) by @qgallouedec in [#49500]
• fix: Pass the MTP layer type to per-layer-type rotary embeddings (#49468) by @shadeMe in [#49468]
• Add LFM2 to the GGUF loading path (#49400) by @francip in [#49400]
• Add MtpModel helpers for frameworks that own the drafting loop (#49544) by @hmellor in [#49544]
• Build docs with the Python version from .python-version (#49603) by @tarekziade in [#49603]
• Use mlp_layer_types for AXK1 so the MoE layer layout honours moe_layer_freq (#49518) by @hmellor in [#49518]
• Combine Nvidia+AMD run-slow CI comments into one (#49598) by @ydshieh in [#49598]
• Fix flaky streamer timeout tests (#49583) by @tarekziade in [#49583]
• Fix TPU related tests issues (#49568) by @tengomucho in [#49568]
• fix(distributed): bind TPU ranks to device index 0, not LOCAL_RANK (#49543) by @tengomucho in [#49543]
• Use transformers init everywhere (#49573) by @guarin in [#49573]
• [docs] Context parallelism during eval (#48663) by @stevhliu in [#48663]
• Remove examples (#49357) by @SunMarc in [#49357]
• Accept whole-number floats for integer tool arguments in parse_response (#49547) by @yonigozlan in [#49547]
• update Intel CPU Dockerfile (#49408) by @jiqing-feng in [#49408]
• Add missing video processor auto docstrings (#49569) by @guarin in [#49569]
• Add chat input sanitization (#47386) by @Rocketknight1 in [#47386]
• Fix create_chunked_causal_mask() crashing with block_sequence_ids (#49457) by @kaixuanliu in [#49457]
• Fix SinglePositionMultiTokenCandidateGenerator signature (#49413) by @jiqing-feng in [#49413]
• Minicpm 4.7 update integration tests (#49565) by @zucchini-nlp in [#49565]
• Stop treating HunYuan's DynamicNTKAlpha RoPE as dynamic (#49445) by @jiqing-feng in [#49445]
• Fix Gemma3nVision2TextModelTest::test_model_parallelism split (#49562) by @tarekziade in [#49562]
• Fix UnboundLocalError in SuperPoint forward under torch.compile (#49409) by @kaixuanliu in [#49409]
• Preserve resolved remote backend references in AutoVideoProcessor (#49511) by @DimensionSTP in [#49511]
• Fix PR slow CI suggestion: install transformers in a venv (#49558) by @tarekziade in [#49558]
• Fix PR slow CI suggestion: install transformers, fail on script errors (#49549) by @ArthurZucker in [#49549]
• Remove deprecated data helpers (#49360) by @SunMarc in [#49360]
• Fix test_all_params_have_gradient for intentionally frozen parameters (#49264) by @thillai-c in [#49264]
• Flex attn fix for molmo2 test (#49530) by @molbap in [#49530]
• fix bridgetower cross-attn mask shape bug (#49447) by @kaixuanliu in [#49447]
• Use Siglip2Tokenizer when the Hub tokenizer_class is GemmaTokenizer (#49245) by @jayzuccarelli in [#49245]
• Fix mixed BOS prefix validation in WatermarkDetector (#49496) by @dangruiyuan in [#49496]
• Deprecate unused get_uniform_frame_indices (#49464) by @archiep867-sys in [#49464]
• [tests] Some fixes for test suite failures on TPU (#49425) by @tengomucho in [#49425]
• Second phase of torch 2.6 cleanups (#49396) by @Rocketknight1 in [#49396]
• Guard the checks for current_accelerator() and bump Torch version (#49395) by @Rocketknight1 in [#49395]
• Use one static Transformers CI badge in PR descriptions (#49471) by @tarekziade in [#49471]
• fix: treat every declared field of the config class as a default fixes #49456 (#49466) by @tarekziade in [#49466]
• Keep AutoVideoProcessor class lookup lazy (#49458) by @DimensionSTP in [#49458]
• Add Molmo2 (#43451) by @SangbumChoi in [#43451]
• Ci fetches multimodal models recursively (#49389) by @zucchini-nlp in [#49389]
• Recurse for all subconfigs (#49461) by @zucchini-nlp in [#49461]
• Aria, Ernie4.5-VL-MoE, Gemma4, Inkling, MiniMax-M3-VL, Qwen3.5-MoE: run the model tests on the VLM/ALM testers (#49404) by @IlyasMoutawwakil in [#49404]
• Fix CI slack reporting: surface crashed/missing model jobs in model_results.json (#49427) by @ydshieh in [#49427]
• [Qwen3VLMoe] Update expected text in test_small_model_integration_test_expand (#49301) by @ydshieh in [#49301]
• Fix backend attribute in documentation and default backend in MistralCommonBackend (#49149) by @samyuktahegde in [#49149]
• More RoPE simplification, TF32 fixes (#49328) by @Rocketknight1 in [#49328]
• Complete PR #49228: remove unused device=None from rope init in 11 model families (#49285) by @ydshieh in [#49285]
• Revert "[CI] Temporarily disable AMD scheduled CI trigger" (#49433) by @ydshieh in [#49433]
• Fix QuantizedCache residual flushing for multi-token updates (#49418) by @LE0-Lin in [#49418]
• Fix data-dependent branch in _ignore_causal_mask_sdpa under compile (#49354) by @dacorvo in [#49354]
• Fix Inkling moe_intermediate_size when dense_intermediate_size is present (#49416) by @Talhax55z in [#49416]
• fix bug for DebertaV2ForMaskedLM non-legacy load issue (#49152) by @kaixuanliu in [#49152]
• This PR fixes Gemma3n fp16 loading when missing weights need to be in… (#49147) by @sywangyi in [#49147]
• PR CI: grant actions: read to the reusable workflow (#49417) by @tarekziade in [#49417]
• [Bugfix] [Modeling] Fix DSV4 norm (#47486) by @kylesayrs in [#47486]
• [docs] Update Spaces (#49242) by @stevhliu in [#49242]
• handle empty precompiled charsmap in spm converter (#49211) by @itazap in [#49211]
• [CI] Temporarily disable AMD scheduled CI trigger (#49394) by @ydshieh in [#49394]
• [tests] Fix flaky CMVN tests in Speech2TextFeatureExtractionWithoutTorchaudioTest (#49374) by @ydshieh in [#49374]
• Let token dispatch run with ep_size != tp_size (#49368) by @qgallouedec in [#49368]
• [tests] Skip onnx.exactness for BitForImageClassification (ORT fp32 drift) (#49366) by @ydshieh in [#49366]

Significant community contributions

The following contributors have made significant changes to the library over the last release:

• @itazap
  • Add MOSS-Transcribe-Diarize model (#48386)
  • handle empty precompiled charsmap in spm converter (#49211)
  • 🚨🚨🚨 luke fixe (#48408)
• @ydshieh
  • Add MIC contract lock files for PR CI integration (#49631)
  • Combine Nvidia+AMD run-slow CI comments into one (#49598)
  • Fix CI slack reporting: surface crashed/missing model jobs in model_results.json (#49427)
  • [Qwen3VLMoe] Update expected text in test_small_model_integration_test_expand (#49301)
  • Complete PR #49228: remove unused device=None from rope init in 11 model families (#49285)
  • Revert "[CI] Temporarily disable AMD scheduled CI trigger" (#49433)
  • [CI] Temporarily disable AMD scheduled CI trigger (#49394)
  • [tests] Fix flaky CMVN tests in Speech2TextFeatureExtractionWithoutTorchaudioTest (#49374)
  • [Docker] Upgrade to torch 2.14/CUDA 13.0 and optimize FA2 build from source (#49381)
  • [tests] Skip onnx.exactness for BitForImageClassification (ORT fp32 drift) (#49366)
• @icavan
  • Fix a port allocation race in FSDP tests (#49455)
  • [Model] Add native Ling 3.0 Flash support (#49446)
• @tarekziade
  • Build docs with the Python version from .python-version (#49603)
  • Fix flaky streamer timeout tests (#49583)
  • CI: Move GPU images to python 3.11 (#49352)
  • Fix Gemma3nVision2TextModelTest::test_model_parallelism split (#49562)
  • Fix PR slow CI suggestion: install transformers in a venv (#49558)
  • Use one static Transformers CI badge in PR descriptions (#49471)
  • fix: treat every declared field of the config class as a default fixes #49456 (#49466)
  • PR CI: grant actions: read to the reusable workflow (#49417)
• @lucamotz
  • Add Kolibri 1 (#49463)
• @SangbumChoi
  • Add Molmo2 (#43451)
• @maxiaolong03
  • [Model] Add PP-DocLayoutV4 Model Support (#48387)

Fine-grained access tokens and service account tokens can now be granted a new contributor permission. A token with it can create repositories in your account or in the organizations you select, and gets full write access to the repositories it created itself. It can't write to any other repository, even ones you or your other tokens created.

This is useful for CI/CD pipelines, training jobs and agents that publish new checkpoints or datasets: they no longer need write access to every repository in the namespace. It works like the contribute-repos scope already available to OAuth apps.