Hugging Face Release Notes

Patch release v5.10.4

Update: Note that on pypi 5.10.3 doesn't exist and this this saved under 5.10.4 (so essentially a minor version skipped). Sorry about that, that's on me. Just wanted to clarify to make this less confusing!

A few fixes needed for vLLM to sync with transformers :hugs:

• [fix] regression introduced by #45534 #46456 by @eustlb (#46456)
• Fix {image/video/audio}_token_ids in ProcessorMixin #46500 by @hmellor (#46500)
• Fix InternVL models #46524 by @hmellor (#46524)
• Fix the offsets in processing #46525 by @zucchini-nlp (#46525)
• Fix peft lower bound #46605 by @hmellor (#46605)
• mistral common backend fix #46667 by @itazap (#46667)

Full Changelog: https://github.com/huggingface/transformers/compare/v5.10.2...v5.10.3

Release v5.12.0

New Model additions

MiniMax-M3-VL

<img width="886" height="583" alt="image" src="https://github.com/user-attachments/assets/ae9dd96f-6877-4531-a06b-a756686f24e5" />

MiniMax-M3-VL is the vision-language member of the MiniMax-M3 family that pairs a CLIP-style vision tower with 3D rotary position embeddings with the MiniMax-M3 text backbone. It uses a mixed dense/sparse Mixture-of-Experts decoder with SwiGLU-OAI gated experts and a lightning indexer for block-sparse attention. The model processes images through a Conv3d patch embedding system and includes specialized components for efficient multimodal understanding and generation.

Links: Documentation

• Add minimax m3vl (#46600) by @ArthurZucker in #46600

PP-OCRv6: update documentation and slow tests (#46576)

<img width="3840" height="1494" alt="image" src="https://github.com/user-attachments/assets/e62284ec-78bf-49cb-8aa2-deccc665372f" /> …

Release v5.11.0

New Model additions

DiffusionGemma

<img width="1240" height="700" alt="image" src="https://github.com/user-attachments/assets/5081e449-6374-4076-bd96-d295c8334ca4" />

DiffusionGemma is engineered to reduce the sequential bottlenecks of standard causal language models by employing an encoder-decoder architecture specifically optimized for inference speed. During inference, DiffusionGemma leverages multi-canvas sampling, where rather than generating one token at a time, the model iteratively denoises a full block of tokens using a diffusion sampler. This block-autoregressive approach facilitates text generation at higher speeds compared to traditional sequential generation methods.

Links: Documentation

• GPU go brr (#46540) by @gante in #46540

DeepSeek-V3.2

<img width="1135" height="671" alt="image" src="https://github.com/user-attachments/assets/24c9694d-eeae-402c-9a98-f7a3971dd9d0" /> …

Release v5.10.1

v5.10.0 was yanked as we publish on a corrupted branch. Sorry everyone, this happens when we rush a release!!!

New Model additions

Gemma4 unified+ Gemma4 MTP

<img width="2000" height="400" alt="image" src="https://github.com/user-attachments/assets/5e3ee940-f78d-4343-ac7a-889930800aa6" />

Gemma 4 12B Unified is an encoder-free multimodal model with pretrained and instruction-tuned variants. Unlike standard Gemma 4, which uses dedicated encoder towers, Gemma 4 12B Unified projects raw inputs directly into the language model's embedding space through lightweight linear pipelines. This results in a simpler architecture while maintaining strong multimodal performance.

Key differences from standard Gemma 4:

• No Vision Tower: Raw pixel patches are projected directly into LM space via a Dense + LayerNorm pipeline with factorized 2D positional embeddings, replacing the vision encoder.
• No Audio Tower: Raw 16 kHz waveform samples are chunked into fixed-length frames and projected through a simple RMSNorm → Linear pipeline, replacing the mel spectrogram + Conformer encoder.
• Shared Multimodal Pipeline: Both vision and audio use the same Gemma4UnifiedMultimodalEmbedder (RMSNorm → Linear) for the final projection to text hidden space.

You can find the original Gemma 4 12B Unified checkpoints under the Gemma 4 release. …

Release v5.9.0

New Model additions

Cohere2Moe

Command A+ is a Mixture-of-Experts (MoE) language model from Cohere that features a hybrid attention pattern combining sliding window and full attention layers. The model incorporates both shared and routed experts and supports a very large context window for processing extensive text sequences.

Links: Documentation

• Add new cohere2_moe model (#46115) by @Cyrilvallez in #46115

Parakeet tdt (#44171)

• Parakeet tdt (#44171) by @lmaksym

HRM-Text

HRM-Text is an improved autoregressive language-modeling variant of the Hierarchical Reasoning Model (HRM) that uses a hierarchical recurrent forward pass with two transformer stacks - one for slow, abstract planning (H) and one for fast, detailed computation (L) - reused inside a nested recurrence. It features PrefixLM attention where instruction tokens attend bidirectionally while response tokens attend causally, per-head sigmoid output gates, and parameterless RMSNorm. The model is designed as a base language model without instruction tuning or chat templates.

Links: Documentation | Paper …

Release v5.8.0

New Model additions

DeepSeek-V4

<img width="6604" height="3574" alt="image" src="https://github.com/user-attachments/assets/4c0fdb29-f770-463c-a97b-d24438896a4c" />

DeepSeek-V4 is the next-generation MoE (Mixture of Experts) language model from DeepSeek that introduces several architectural innovations over DeepSeek-V3. The architecture replaces Multi-head Latent Attention (MLA) with a hybrid local + long-range attention design, swaps residual connections for Manifold-Constrained Hyper-Connections (mHC), and bootstraps the first few MoE layers with a static token-id → expert-id hash table. This implementation covers DeepSeek-V4-Flash, DeepSeek-V4-Pro, and their -Base pretrained variants, which share the same architecture but differ in width, depth, expert count and weights.

Links: Documentation | Paper

• Add DeepSeek V4 (#45643) by @ArthurZucker in #45643

Gemma 4 Assistant

<img width="2000" height="400" alt="image" src="https://github.com/user-attachments/assets/02c79b0b-a172-4495-b09d-a6a4b625ee66" /> …

New Pipelines

LLaDA2

LLaDA2 is a family of discrete diffusion language models that generate text through block-wise iterative refinement. Instead of autoregressive token-by-token generation, LLaDA2 starts with a fully masked sequence and progressively unmasks tokens by confidence over multiple refinement steps.

• PR: https://github.com/huggingface/diffusers/pull/13226
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/llada2

Nucleus-MoE

NucleusMoE-Image is a 2B active 17B parameter model trained with efficiency at its core. Our novel architecture highlights the scalability of a sparse MoE architecture for Image generation.

• PR: https://github.com/huggingface/diffusers/pull/13317
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/nucleusmoe_image

Thanks to @sippycoder for the contribution.

Ernie-Image

ERNIE-Image is a powerful and highly efficient image generation model with 8B parameters.

• PR: https://github.com/huggingface/diffusers/pull/13432
• Docs: https://huggingface.co/docs/diffusers/main/api/pipelines/ernie_image

Thanks to @HsiaWinter for the contribution.

LongCat-AudioDiT …

Release v5.7.0

New Model additions

Laguna

<img width="699" height="176" alt="image" src="https://github.com/user-attachments/assets/d3bae269-bea7-4ddf-a53f-d4718befdb17" />

Laguna is Poolside's mixture-of-experts language model family that extends standard SwiGLU MoE transformers with two key innovations. It features per-layer head counts allowing different decoder layers to have different query-head counts while sharing the same KV cache shape, and implements a sigmoid MoE router with auxiliary-loss-free load balancing that uses element-wise sigmoid of gate logits plus learned per-expert bias for router scoring.

Links: Documentation

• Laguna XS.2 implementation (#45673) by @joerowell in #45673

DEIMv2

<img width="2874" height="908" alt="image" src="https://github.com/user-attachments/assets/fc8c59fe-f964-42ce-ae8e-c7fcace9beb7" /> …

Release v5.6.0

New Model additions

OpenAI Privacy Filter

OpenAI Privacy Filter is a bidirectional token-classification model for personally identifiable information (PII) detection and masking in text. It is intended for high-throughput data sanitization workflows where teams need a model that they can run on-premises that is fast, context-aware, and tunable. The model labels an input sequence in a single forward pass, then decodes coherent spans with a constrained Viterbi procedure, predicting probability distributions over 8 privacy-related output categories for each input token.

Links: Documentation

• [Privacy Filter] Add model (#45580) by @vasqu in #45580

QianfanOCR

Qianfan-OCR is a 4B-parameter end-to-end document intelligence model developed by Baidu that performs direct image-to-text conversion without traditional multi-stage OCR pipelines. It supports a broad range of prompt-driven tasks including structured document parsing, table extraction, chart understanding, document question answering, and key information extraction all within one unified model. The model features a unique "Layout-as-Thought" capability that generates structured layout representations before producing final outputs, making it particularly effective for complex documents with mixed element types. …

Release v5.5.0

<img width="2786" height="1504" alt="image" src="https://github.com/user-attachments/assets/6c8c878f-042b-4858-9f64-73fd9ccd7e4b" />

New Model additions

Gemma4

Gemma 4 is a multimodal model with pretrained and instruction-tuned variants, available in 1B, 13B, and 27B parameters. The architecture is mostly the same as the previous Gemma versions. The key differences are a vision processor that can output images of fixed token budget and a spatial 2D RoPE to encode vision-specific information across height and width axis.

<img width="1478" height="1374" alt="image" src="https://github.com/user-attachments/assets/9d88bd1b-02ea-4829-b7d0-fac0e347d436" />

You can find all the original Gemma 4 checkpoints under the Gemma 4 release.

The key difference from previous Gemma releases is the new design to process images of different sizes using a fixed-budget number of tokens. Unlike many models that squash every image into a fixed square (like 224×224), Gemma 4 keeps the image's natural aspect ratio while making it the right size. There a a couple constraints to follow:

• The total number of pixels must fit within a patch budget
• Both height and width must be divisible by 48 (= patch size 16 × pooling kernel 3)

[!IMPORTANT] …

New Model additions

VidEoMT

<img width="1480" height="460" alt="image" src="https://github.com/user-attachments/assets/bec6fc25-b0ab-4227-8c2b-a838554f37f3" />

Video Encoder-only Mask Transformer (VidEoMT) is a lightweight encoder-only model for online video segmentation built on a plain Vision Transformer (ViT). It eliminates the need for dedicated tracking modules by introducing a lightweight query propagation mechanism that carries information across frames and employs a query fusion strategy that combines propagated queries with temporally-agnostic learned queries. VidEoMT achieves competitive accuracy while being 5x-10x faster than existing approaches, running at up to 160 FPS with a ViT-L backbone.

Links: Documentation | Paper

• Add VidEoMT (#44285) by @NielsRogge in #44285

UVDoc

<img width="1765" height="875" alt="image" src="https://github.com/user-attachments/assets/365e510e-8fb8-46cb-8f4b-e8b7082f0ae2" /> …