renovate-bot opened a new pull request, #8313: URL: https://github.com/apache/texera/pull/8313
> ℹ️ **Note** > > This PR body was truncated due to platform limits. This PR contains the following updates: | Package | Change | [Age](https://docs.renovatebot.com/merge-confidence/) | [Confidence](https://docs.renovatebot.com/merge-confidence/) | |---|---|---|---| | [transformers](https://redirect.github.com/huggingface/transformers) | `==5.5.0` → `==5.10.1` |  |  | --- ### Transformers save_pretrained path traversal allows arbitrary file writes through chat template names [CVE-2026-9856](https://nvd.nist.gov/vuln/detail/CVE-2026-9856) / [GHSA-xrqw-3rrv-vx5w](https://redirect.github.com/advisories/GHSA-xrqw-3rrv-vx5w) <details> <summary>More information</summary> #### Details A vulnerability in huggingface/transformers versions < 5.10.0 allows an attacker to perform arbitrary file writes via path traversal. The issue resides in the `save_pretrained()` methods of `PreTrainedTokenizerBase` and `ProcessorMixin`, where keys from the `chat_template` dictionary are used directly as filenames without proper validation. An attacker can exploit this by publishing a malicious Hugging Face Hub repository with a crafted `tokenizer_config.json` file. When a victim downloads and saves the tokenizer or processor, the attacker-controlled keys can escape the intended save directory, enabling arbitrary file writes with attacker-controlled content. This vulnerability affects multiple processors inheriting from `ProcessorMixin`, including Idefics, Florence, Gemma, Phi, and Qwen-VL. #### Severity - CVSS Score: 7.1 / 10 (High) - Vector String: `CVSS:3.0/AV:N/AC:L/PR:N/UI:R/S:U/C:N/I:H/A:L` #### References - [https://nvd.nist.gov/vuln/detail/CVE-2026-9856](https://nvd.nist.gov/vuln/detail/CVE-2026-9856) - [https://github.com/huggingface/transformers/commit/eaaaf8494dd5386634ae37d1d122212fdc315be5](https://redirect.github.com/huggingface/transformers/commit/eaaaf8494dd5386634ae37d1d122212fdc315be5) - [https://huntr.com/bounties/362824d5-fe18-40e8-a6cf-62277f97a170](https://huntr.com/bounties/362824d5-fe18-40e8-a6cf-62277f97a170) - [https://github.com/huggingface/transformers/pull/46191](https://redirect.github.com/huggingface/transformers/pull/46191) - [https://github.com/advisories/GHSA-xrqw-3rrv-vx5w](https://redirect.github.com/advisories/GHSA-xrqw-3rrv-vx5w) This data is provided by the [GitHub Advisory Database](https://redirect.github.com/advisories/GHSA-xrqw-3rrv-vx5w) ([CC-BY 4.0](https://redirect.github.com/github/advisory-database/blob/main/LICENSE.md)). </details> --- ### Release Notes <details> <summary>huggingface/transformers (transformers)</summary> ### [`v5.10.1`](https://redirect.github.com/huggingface/transformers/releases/tag/v5.10.1) [Compare Source](https://redirect.github.com/huggingface/transformers/compare/v5.10.0...v5.10.1) ### 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](./gemma4), 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](https://huggingface.co/collections/google/gemma-4) release. - who needs encoders? ([#​46385](https://redirect.github.com/huggingface/transformers/issues/46385)) by [@​douglas-reid](https://redirect.github.com/douglas-reid) [@​sgerrard](https://redirect.github.com/sgerrard) [@​vasqu](https://redirect.github.com/vasqu) [@​molbap](https://redirect.github.com/molbap) ##### Sapiens2 Sapiens2 is a family of high-resolution vision transformers pretrained on \~1 billion curated human images, designed for human-centric computer vision tasks including pose estimation, body-part segmentation, surface normal estimation, and pointmap estimation. The models scale from 0.4B to 5B parameters and train at native 1K resolution, with hierarchical 4K variants for extended spatial reasoning. Sapiens2 achieves substantial improvements over its predecessor with +4 mAP in pose estimation, +24.3 mIoU in body-part segmentation, and 45.6% error reduction in normal estimation. **Links:** [Documentation](https://huggingface.co/docs/transformers/main/en/model_doc/sapiens2) | [Paper](https://huggingface.co/papers/2604.21681) - Add Sapiens2 Model ([#​45919](https://redirect.github.com/huggingface/transformers/issues/45919)) by [@​guarin](https://redirect.github.com/guarin) in [#​45919](https://redirect.github.com/huggingface/transformers/pull/45919) ##### DeepSeek-OCR-2 DeepSeek-OCR-2 is an OCR-specialized vision-language model built on a distinctive architecture that combines a SAM ViT-B vision encoder with a Qwen2 hybrid attention encoder, connected through an MLP projector to a DeepSeek-V2 Mixture-of-Experts (MoE) language model. The model features a hybrid attention mechanism that applies bidirectional attention over image tokens and causal attention over query tokens, enabling efficient and accurate document understanding. It supports both plain OCR tasks and grounding capabilities with coordinate-aware output for document conversion to markdown format. **Links:** [Documentation](https://huggingface.co/docs/transformers/main/en/model_doc/deepseek_ocr2) - Add Deepseek-OCR-2 model ([#​45075](https://redirect.github.com/huggingface/transformers/issues/45075)) by [@​thisisiron](https://redirect.github.com/thisisiron) in [#​45075](https://redirect.github.com/huggingface/transformers/pull/45075) ##### Mellum Mellum is a code-focused Mixture-of-Experts language model developed by JetBrains. It is derived from the Qwen3-MoE architecture with per-layer-type RoPE and interleaved sliding window attention. The model has 12B total parameters with 2.5B active parameters per token, using 64 routed experts with 8 activated per token across 28 layers. **Links:** [Documentation](https://huggingface.co/docs/transformers/main/en/model_doc/mellum) - feat: Add support for JetBrains' `Mellum` v2 code generation model ([#​46112](https://redirect.github.com/huggingface/transformers/issues/46112)) by [@​shadeMe](https://redirect.github.com/shadeMe) in [#​46112](https://redirect.github.com/huggingface/transformers/pull/46112) #### Breaking changes The Gemma4 vision pooler now casts inputs to float32 before scaling to prevent float16 overflow (inf saturation) with large checkpoints, which may cause minor numerical differences in outputs for users running Gemma-4 vision models in float16. - 🚨 Fix float16 overflow in Gemma4 vision pooler ([#​46277](https://redirect.github.com/huggingface/transformers/issues/46277)) by [@​Bluear7878](https://redirect.github.com/Bluear7878) Audio Language Models (ALMs) now have a dedicated base model class without a language modeling head, aligning them with the design of Vision Language Models (VLMs); users relying on the previous model class structure should update their code to use the new base model class where appropriate. - 🚨 \[ALM] Add base model without head ([#​45534](https://redirect.github.com/huggingface/transformers/issues/45534)) by [@​eustlb](https://redirect.github.com/eustlb) #### Parallelization This release includes numerous bug fixes for model parallelism across multiple models (Gemma4, AltCLIP, ChineseClip, Blip-2, Whisper, Ovis2, Moshi) and parallel execution strategies, including fixes for tensor parallelism (TP), expert parallelism (EP), beam search under model parallel settings, and loss over-counting under TP/EP configurations. The continuous batching manager was also reworked for clearer control flow and improved TP race condition handling, and FSDP initialization via `from_pretrained` was introduced. - Fix dsv4 dequant + tp/ep ([#​46378](https://redirect.github.com/huggingface/transformers/issues/46378)) by [@​IlyasMoutawwakil](https://redirect.github.com/IlyasMoutawwakil) in \[[#​46378](https://redirect.github.com/huggingface/transformers/issues/46378)] - \[CB] \[Major] Rework manager to have clearer control flow + handle TP ([#​46070](https://redirect.github.com/huggingface/transformers/issues/46070)) by [@​remi-or](https://redirect.github.com/remi-or) in \[[#​46070](https://redirect.github.com/huggingface/transformers/issues/46070)] - fix series of bugs for model parallel beam search ([#​46280](https://redirect.github.com/huggingface/transformers/issues/46280)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46280](https://redirect.github.com/huggingface/transformers/issues/46280)] - Fix model parallel issue for altclip model and ChineseClip model ([#​45487](https://redirect.github.com/huggingface/transformers/issues/45487)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​45487](https://redirect.github.com/huggingface/transformers/issues/45487)] - Model parallel fix ([#​46230](https://redirect.github.com/huggingface/transformers/issues/46230)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46230](https://redirect.github.com/huggingface/transformers/issues/46230)] - \[`Revert`] FSDP+Dtensor refactor related changes ([#​46246](https://redirect.github.com/huggingface/transformers/issues/46246)) by [@​vasqu](https://redirect.github.com/vasqu) in \[[#​46246](https://redirect.github.com/huggingface/transformers/issues/46246)] - Fix model parallel bugs for Gemma4 ([#​45817](https://redirect.github.com/huggingface/transformers/issues/45817)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​45817](https://redirect.github.com/huggingface/transformers/issues/45817)] - init FSDP through from\_pretrained ([#​46102](https://redirect.github.com/huggingface/transformers/issues/46102)) by [@​3outeille](https://redirect.github.com/3outeille) in \[[#​46102](https://redirect.github.com/huggingface/transformers/issues/46102)] - fix model parallel device mismatch issue in `create_bidirectional_mask` ([#​46221](https://redirect.github.com/huggingface/transformers/issues/46221)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46221](https://redirect.github.com/huggingface/transformers/issues/46221)] - Trainer.compute\_loss: fix loss over-counting under TP and EP-as-TP ([#​45994](https://redirect.github.com/huggingface/transformers/issues/45994)) by [@​AmineDiro](https://redirect.github.com/AmineDiro) in \[[#​45994](https://redirect.github.com/huggingface/transformers/issues/45994)] - Fix caching allocator warmup byte estimation for EP model loading ([#​46149](https://redirect.github.com/huggingface/transformers/issues/46149)) by [@​sywangyi](https://redirect.github.com/sywangyi) in \[[#​46149](https://redirect.github.com/huggingface/transformers/issues/46149)] #### Cache Fixed a regression in encoder-decoder cache initialization where the decoder config was incorrectly applied to the cross-attention cache, and resolved a `RuntimeError` caused by buffer size limits when warming up the cache on MPS devices. Additional test infrastructure improvements were made to support read-only cache environments used in CI. - fix: cache warmup `RuntimeError` on mps ([#​46239](https://redirect.github.com/huggingface/transformers/issues/46239)) by [@​McPatate](https://redirect.github.com/McPatate) in \[[#​46239](https://redirect.github.com/huggingface/transformers/issues/46239)] - Make more tests work with read-only cache ([#​46299](https://redirect.github.com/huggingface/transformers/issues/46299)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46299](https://redirect.github.com/huggingface/transformers/issues/46299)] - Update a test to avoid writing to the default xet cache ([#​46250](https://redirect.github.com/huggingface/transformers/issues/46250)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46250](https://redirect.github.com/huggingface/transformers/issues/46250)] - Fix a regression in encoder-decoder generation cache initialization ([#​46111](https://redirect.github.com/huggingface/transformers/issues/46111)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46111](https://redirect.github.com/huggingface/transformers/issues/46111)] #### Quantization Added support for DeepGEMM BF16, mixed FP8/FP4, and MegaMoE quantization via a grouped linear refactor, while fixing two bugs: an FP8 MoE reverse substring issue affecting DSv4 initialization, and a BitsAndBytes 4-bit/8-bit quantization bug that silently dropped chunked tensors from one-to-many weight converters. - DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor ([#​45634](https://redirect.github.com/huggingface/transformers/issues/45634)) by [@​IlyasMoutawwakil](https://redirect.github.com/IlyasMoutawwakil) in \[[#​45634](https://redirect.github.com/huggingface/transformers/issues/45634)] - Fix fp8 moe reverse substring ([#​46265](https://redirect.github.com/huggingface/transformers/issues/46265)) by [@​ArthurZucker](https://redirect.github.com/ArthurZucker) in \[[#​46265](https://redirect.github.com/huggingface/transformers/issues/46265)] - Fix bnb 4bit/8bit quantization drop chunked tensors bug ([#​46210](https://redirect.github.com/huggingface/transformers/issues/46210)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46210](https://redirect.github.com/huggingface/transformers/issues/46210)] #### Bugfixes and improvements - Fix wrong changes produced by style/repo. check bot ([#​46371](https://redirect.github.com/huggingface/transformers/issues/46371)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46371](https://redirect.github.com/huggingface/transformers/issues/46371)] - Fix path traversal when saving Bark voice preset embeddings ([#​46237](https://redirect.github.com/huggingface/transformers/issues/46237)) by [@​LinZiyuu](https://redirect.github.com/LinZiyuu) in \[[#​46237](https://redirect.github.com/huggingface/transformers/issues/46237)] - Pass library\_name/version to Hub calls via a shared HfApi ([#​46318](https://redirect.github.com/huggingface/transformers/issues/46318)) by [@​Wauplin](https://redirect.github.com/Wauplin) in \[[#​46318](https://redirect.github.com/huggingface/transformers/issues/46318)] - docs: update ACL Anthology URL in CITATION.cff ([#​46352](https://redirect.github.com/huggingface/transformers/issues/46352)) by [@​irfaan101](https://redirect.github.com/irfaan101) in \[[#​46352](https://redirect.github.com/huggingface/transformers/issues/46352)] - \[docs] contributing ([#​45465](https://redirect.github.com/huggingface/transformers/issues/45465)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​45465](https://redirect.github.com/huggingface/transformers/issues/45465)] - \[docs] Romanian translation of `contributing.md`, `modular_transformers.md`, `multimodal_processing.md`, `add_vision_processing_components.md`, `add_audio_processing_components.md`, `modeling_rules.md`, `model_output_tracing.md`, `auto_docstring.md`, `testing.md`, `pr_checks.md` and `add_new_model.md` . ([#​46345](https://redirect.github.com/huggingface/transformers/issues/46345)) by [@​filipinescu](https://redirect.github.com/filipinescu) in \[[#​46345](https://redirect.github.com/huggingface/transformers/issues/46345)] - \[docs] xpu continuous batching ([#​46334](https://redirect.github.com/huggingface/transformers/issues/46334)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​46334](https://redirect.github.com/huggingface/transformers/issues/46334)] - Fix incorrect attribute mapping relationships in GLM MoE DSA Config ([#​46338](https://redirect.github.com/huggingface/transformers/issues/46338)) by [@​Dovis01](https://redirect.github.com/Dovis01) in \[[#​46338](https://redirect.github.com/huggingface/transformers/issues/46338)] - Fix grammar typos in Whisper documentation ([#​46336](https://redirect.github.com/huggingface/transformers/issues/46336)) by [@​calliec-1223](https://redirect.github.com/calliec-1223) in \[[#​46336](https://redirect.github.com/huggingface/transformers/issues/46336)] - \[docs] update num\_items\_in\_batch for causal LMs ([#​46335](https://redirect.github.com/huggingface/transformers/issues/46335)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​46335](https://redirect.github.com/huggingface/transformers/issues/46335)] - Update compressed tensors minimum version ([#​46342](https://redirect.github.com/huggingface/transformers/issues/46342)) by [@​SunMarc](https://redirect.github.com/SunMarc) in \[[#​46342](https://redirect.github.com/huggingface/transformers/issues/46342)] - Fix \_is\_package\_available reporting available without a version ([#​46125](https://redirect.github.com/huggingface/transformers/issues/46125)) by [@​blipbyte](https://redirect.github.com/blipbyte) in \[[#​46125](https://redirect.github.com/huggingface/transformers/issues/46125)] - remove sec ([#​46346](https://redirect.github.com/huggingface/transformers/issues/46346)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46346](https://redirect.github.com/huggingface/transformers/issues/46346)] - fix: include transitive relative imports when loading from local directory ([#​46022](https://redirect.github.com/huggingface/transformers/issues/46022)) by [@​trducng](https://redirect.github.com/trducng) in \[[#​46022](https://redirect.github.com/huggingface/transformers/issues/46022)] - perf(feature\_extraction\_sequence): skip re-splitting already-batched numpy arrays in pad() ([#​46329](https://redirect.github.com/huggingface/transformers/issues/46329)) by [@​Anai-Guo](https://redirect.github.com/Anai-Guo) in \[[#​46329](https://redirect.github.com/huggingface/transformers/issues/46329)] - \[Zamba] Support attn\_implementation dispatch ([#​46317](https://redirect.github.com/huggingface/transformers/issues/46317)) by [@​YangKai0616](https://redirect.github.com/YangKai0616) in \[[#​46317](https://redirect.github.com/huggingface/transformers/issues/46317)] - Fix TestAppRoutes test failures caused by deprecated asyncio.get\_event\_loop() on Python 3.10+ ([#​46340](https://redirect.github.com/huggingface/transformers/issues/46340)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46340](https://redirect.github.com/huggingface/transformers/issues/46340)] - \[Qwen3VL] Fix video token placeholder: use self.video\_token instead of hardcoded "<|placeholder|>" ([#​46296](https://redirect.github.com/huggingface/transformers/issues/46296)) by [@​kpal002](https://redirect.github.com/kpal002) in \[[#​46296](https://redirect.github.com/huggingface/transformers/issues/46296)] - chore(linter): fixes for rule 16 ([#​46023](https://redirect.github.com/huggingface/transformers/issues/46023)) by [@​tarekziade](https://redirect.github.com/tarekziade) in \[[#​46023](https://redirect.github.com/huggingface/transformers/issues/46023)] - \[docs] Romanian translation of `weightconverter.md`, `models.md`, `custom_models.md`, `monkey_patching.md`, `fusion_mapping.md`, `how_to_hack_models.md`, `model_sharing.md` and `serialization.md`. ([#​46309](https://redirect.github.com/huggingface/transformers/issues/46309)) by [@​filipinescu](https://redirect.github.com/filipinescu) in \[[#​46309](https://redirect.github.com/huggingface/transformers/issues/46309)] - Normalize CUDA OOM errors when comparing commit failures in check\_bad\_commit ([#​46322](https://redirect.github.com/huggingface/transformers/issues/46322)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46322](https://redirect.github.com/huggingface/transformers/issues/46322)] - Fix unhandled exception noise from background safetensors conversion thread ([#​45752](https://redirect.github.com/huggingface/transformers/issues/45752)) by [@​dhruv7477](https://redirect.github.com/dhruv7477) in \[[#​45752](https://redirect.github.com/huggingface/transformers/issues/45752)] - Add Expectations for pipeline token classification tests ([#​46151](https://redirect.github.com/huggingface/transformers/issues/46151)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46151](https://redirect.github.com/huggingface/transformers/issues/46151)] - \[docs] fix auto-add release dates ([#​46283](https://redirect.github.com/huggingface/transformers/issues/46283)) by [@​zucchini-nlp](https://redirect.github.com/zucchini-nlp) in \[[#​46283](https://redirect.github.com/huggingface/transformers/issues/46283)] - Separate pip command syntax for notebook and CLI tabs in Quickstart ([#​46243](https://redirect.github.com/huggingface/transformers/issues/46243)) by [@​pvelayudhan](https://redirect.github.com/pvelayudhan) in \[[#​46243](https://redirect.github.com/huggingface/transformers/issues/46243)] - Romanian translation of README.md, index.md, installation.md, \_config.py and quicktour.md. ([#​46166](https://redirect.github.com/huggingface/transformers/issues/46166)) by [@​filipinescu](https://redirect.github.com/filipinescu) in \[[#​46166](https://redirect.github.com/huggingface/transformers/issues/46166)] - Fall back to flat kwarg when modality dict is passed without it ([#​46195](https://redirect.github.com/huggingface/transformers/issues/46195)) by [@​Ace3Z](https://redirect.github.com/Ace3Z) in \[[#​46195](https://redirect.github.com/huggingface/transformers/issues/46195)] - Fix load\_adapter OOM caused by full-model warmup sizing ([#​46145](https://redirect.github.com/huggingface/transformers/issues/46145)) by [@​Yooniel](https://redirect.github.com/Yooniel) in \[[#​46145](https://redirect.github.com/huggingface/transformers/issues/46145)] - Replace assert with raise ImportError for optuna/ray dependency checks ([#​46263](https://redirect.github.com/huggingface/transformers/issues/46263)) by [@​SebTardif](https://redirect.github.com/SebTardif) in \[[#​46263](https://redirect.github.com/huggingface/transformers/issues/46263)] - chore(linter): respect TRF017 modeling rule ([#​46260](https://redirect.github.com/huggingface/transformers/issues/46260)) by [@​tarekziade](https://redirect.github.com/tarekziade) in \[[#​46260](https://redirect.github.com/huggingface/transformers/issues/46260)] - Delete dead code in qwen-vl series ([#​45827](https://redirect.github.com/huggingface/transformers/issues/45827)) by [@​zucchini-nlp](https://redirect.github.com/zucchini-nlp) in \[[#​45827](https://redirect.github.com/huggingface/transformers/issues/45827)] - qa: fix ty caching and align CI with local run ([#​46278](https://redirect.github.com/huggingface/transformers/issues/46278)) by [@​tarekziade](https://redirect.github.com/tarekziade) in \[[#​46278](https://redirect.github.com/huggingface/transformers/issues/46278)] - Guard DeviceMesh import in continuous batching ([#​46205](https://redirect.github.com/huggingface/transformers/issues/46205)) by [@​danyalahmed1995](https://redirect.github.com/danyalahmed1995) in \[[#​46205](https://redirect.github.com/huggingface/transformers/issues/46205)] - Processor compatibility with vLLM ([#​46258](https://redirect.github.com/huggingface/transformers/issues/46258)) by [@​zucchini-nlp](https://redirect.github.com/zucchini-nlp) in \[[#​46258](https://redirect.github.com/huggingface/transformers/issues/46258)] - Fix PR CI workflow cancellation condition ([#​46276](https://redirect.github.com/huggingface/transformers/issues/46276)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46276](https://redirect.github.com/huggingface/transformers/issues/46276)] - \[fix] toctree ([#​46106](https://redirect.github.com/huggingface/transformers/issues/46106)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​46106](https://redirect.github.com/huggingface/transformers/issues/46106)] - add more generic support for distributed trainer tests ([#​46109](https://redirect.github.com/huggingface/transformers/issues/46109)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46109](https://redirect.github.com/huggingface/transformers/issues/46109)] - add XPU Expectations for florence2 and lfm2\_vl model test ([#​46275](https://redirect.github.com/huggingface/transformers/issues/46275)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46275](https://redirect.github.com/huggingface/transformers/issues/46275)] - Fix `StaticCache` building an empty layer list when `num_kv_shared_layers == 0` ([#​46235](https://redirect.github.com/huggingface/transformers/issues/46235)) by [@​tengomucho](https://redirect.github.com/tengomucho) in \[[#​46235](https://redirect.github.com/huggingface/transformers/issues/46235)] - Fix inverted assertion in remove\_handler ([#​46227](https://redirect.github.com/huggingface/transformers/issues/46227)) by [@​SebTardif](https://redirect.github.com/SebTardif) in \[[#​46227](https://redirect.github.com/huggingface/transformers/issues/46227)] - \[ShieldGemma2] Support attn\_implementation dispatch ([#​46069](https://redirect.github.com/huggingface/transformers/issues/46069)) by [@​YangKai0616](https://redirect.github.com/YangKai0616) in \[[#​46069](https://redirect.github.com/huggingface/transformers/issues/46069)] - \[Gemma4] Replace one-hot matmul with F.embedding in position embeddings ([#​46176](https://redirect.github.com/huggingface/transformers/issues/46176)) by [@​Sriniketh24](https://redirect.github.com/Sriniketh24) in \[[#​46176](https://redirect.github.com/huggingface/transformers/issues/46176)] - fix: kosmos2.5: properly expand embeddings table ([#​45835](https://redirect.github.com/huggingface/transformers/issues/45835)) by [@​nunq](https://redirect.github.com/nunq) in \[[#​45835](https://redirect.github.com/huggingface/transformers/issues/45835)] - find pytest launch error in torch 2.13.0.dev20260526 ([#​46252](https://redirect.github.com/huggingface/transformers/issues/46252)) by [@​sywangyi](https://redirect.github.com/sywangyi) in \[[#​46252](https://redirect.github.com/huggingface/transformers/issues/46252)] - \[Test]\[Kosmos2.5] Add XPU expectations for integration tests ([#​46135](https://redirect.github.com/huggingface/transformers/issues/46135)) by [@​YangKai0616](https://redirect.github.com/YangKai0616) in \[[#​46135](https://redirect.github.com/huggingface/transformers/issues/46135)] - Support FA2 flash\_attn\_with\_kvcache for XPU continuous batching ([#​46028](https://redirect.github.com/huggingface/transformers/issues/46028)) by [@​YangKai0616](https://redirect.github.com/YangKai0616) in \[[#​46028](https://redirect.github.com/huggingface/transformers/issues/46028)] - \[`Configs`] Fix layer type validation to include its mlp counterpart ([#​46220](https://redirect.github.com/huggingface/transformers/issues/46220)) by [@​vasqu](https://redirect.github.com/vasqu) in \[[#​46220](https://redirect.github.com/huggingface/transformers/issues/46220)] - Fix `num_items_in_batch` over-counting for causal LM losses ([#​46204](https://redirect.github.com/huggingface/transformers/issues/46204)) by [@​qgallouedec](https://redirect.github.com/qgallouedec) in \[[#​46204](https://redirect.github.com/huggingface/transformers/issues/46204)] - RF-DETR doc fixes ([#​46244](https://redirect.github.com/huggingface/transformers/issues/46244)) by [@​merveenoyan](https://redirect.github.com/merveenoyan) in \[[#​46244](https://redirect.github.com/huggingface/transformers/issues/46244)] - Use `main` instead of commit SHA for now ([#​46241](https://redirect.github.com/huggingface/transformers/issues/46241)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46241](https://redirect.github.com/huggingface/transformers/issues/46241)] - Enable push event (to main) for PR CI workflow ([#​46240](https://redirect.github.com/huggingface/transformers/issues/46240)) by [@​ydshieh](https://redirect.github.com/ydshieh) in \[[#​46240](https://redirect.github.com/huggingface/transformers/issues/46240)] - fix(hrm\_text): Add XPU Expectations for tests ([#​46214](https://redirect.github.com/huggingface/transformers/issues/46214)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​46214](https://redirect.github.com/huggingface/transformers/issues/46214)] - \[deepseek\_v4] keep hc\_head / sinks / position\_bias in fp32 ([#​46198](https://redirect.github.com/huggingface/transformers/issues/46198)) by [@​ArthurZucker](https://redirect.github.com/ArthurZucker) in \[[#​46198](https://redirect.github.com/huggingface/transformers/issues/46198)] - Fix FSDP2 and distributed checkpointing imports for older PyTorch versions ([#​46141](https://redirect.github.com/huggingface/transformers/issues/46141)) by [@​ryota-komatsu](https://redirect.github.com/ryota-komatsu) in \[[#​46141](https://redirect.github.com/huggingface/transformers/issues/46141)] - Fix Gemma4 Array Mask Indexing ([#​46203](https://redirect.github.com/huggingface/transformers/issues/46203)) by [@​petecao](https://redirect.github.com/petecao) in \[[#​46203](https://redirect.github.com/huggingface/transformers/issues/46203)] - utils: handle flash\_attn missing from importlib packages\_distributions without crashing ([#​45524](https://redirect.github.com/huggingface/transformers/issues/45524)) by [@​SAY-5](https://redirect.github.com/SAY-5) in \[[#​45524](https://redirect.github.com/huggingface/transformers/issues/45524)] - \[AMD CI] revert AMD mi325 hf-workflows ref from SHA back to [@​main](https://redirect.github.com/main) ([#​46213](https://redirect.github.com/huggingface/transformers/issues/46213)) by [@​Abdennacer-Badaoui](https://redirect.github.com/Abdennacer-Badaoui) in \[[#​46213](https://redirect.github.com/huggingface/transformers/issues/46213)] - \[GLM-4.6V] Update with GLM-GA Processor ([#​46184](https://redirect.github.com/huggingface/transformers/issues/46184)) by [@​zRzRzRzRzRzRzR](https://redirect.github.com/zRzRzRzRzRzRzR) in \[[#​46184](https://redirect.github.com/huggingface/transformers/issues/46184)] - update xpu expectation for falcon mamba ([#​46086](https://redirect.github.com/huggingface/transformers/issues/46086)) by [@​sywangyi](https://redirect.github.com/sywangyi) in \[[#​46086](https://redirect.github.com/huggingface/transformers/issues/46086)] - chore: enable Dependabot weekly GitHub Actions bumps ([#​46157](https://redirect.github.com/huggingface/transformers/issues/46157)) by [@​hf-dependantbot-rollout](https://redirect.github.com/hf-dependantbot-rollout)\[bot] in \[[#​46157](https://redirect.github.com/huggingface/transformers/issues/46157)] - Fix Gemma4 use\_bidirectional\_attention="all" mask behavior ([#​46079](https://redirect.github.com/huggingface/transformers/issues/46079)) by [@​oliverholworthy](https://redirect.github.com/oliverholworthy) in \[[#​46079](https://redirect.github.com/huggingface/transformers/issues/46079)] - Fix loading with only 1 device or distributed config ([#​46197](https://redirect.github.com/huggingface/transformers/issues/46197)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46197](https://redirect.github.com/huggingface/transformers/issues/46197)] - Fix TypeError on list-typed ignore\_keys\_at\_rope\_validation in RoPE config ([#​46142](https://redirect.github.com/huggingface/transformers/issues/46142)) by [@​Charly21r](https://redirect.github.com/Charly21r) in \[[#​46142](https://redirect.github.com/huggingface/transformers/issues/46142)] - Support XPU autocast dtype fallback for FlashAttention ([#​46199](https://redirect.github.com/huggingface/transformers/issues/46199)) by [@​YangKai0616](https://redirect.github.com/YangKai0616) in \[[#​46199](https://redirect.github.com/huggingface/transformers/issues/46199)] - Fix path traversal when saving named chat templates ([#​46191](https://redirect.github.com/huggingface/transformers/issues/46191)) by [@​LinZiyuu](https://redirect.github.com/LinZiyuu) in \[[#​46191](https://redirect.github.com/huggingface/transformers/issues/46191)] - Fix is\_last off-by-one in MaskGenerationPipeline for partial batches ([#​46136](https://redirect.github.com/huggingface/transformers/issues/46136)) by [@​J3r3myPerera](https://redirect.github.com/J3r3myPerera) in \[[#​46136](https://redirect.github.com/huggingface/transformers/issues/46136)] - Fix wrong variable in check\_model\_type isinstance check ([#​46080](https://redirect.github.com/huggingface/transformers/issues/46080)) by [@​SebTardif](https://redirect.github.com/SebTardif) in \[[#​46080](https://redirect.github.com/huggingface/transformers/issues/46080)] - Enable passing kwargs through RoFormer models ([#​46171](https://redirect.github.com/huggingface/transformers/issues/46171)) by [@​ir2718](https://redirect.github.com/ir2718) in \[[#​46171](https://redirect.github.com/huggingface/transformers/issues/46171)] - Update cohere2\_moe tp\_plan ([#​46189](https://redirect.github.com/huggingface/transformers/issues/46189)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46189](https://redirect.github.com/huggingface/transformers/issues/46189)] - Update release tool ([#​46193](https://redirect.github.com/huggingface/transformers/issues/46193)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46193](https://redirect.github.com/huggingface/transformers/issues/46193)] - \[loading] Fix base\_model\_prefix issues in conversions ([#​46067](https://redirect.github.com/huggingface/transformers/issues/46067)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46067](https://redirect.github.com/huggingface/transformers/issues/46067)] - Bump dev version ([#​46188](https://redirect.github.com/huggingface/transformers/issues/46188)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46188](https://redirect.github.com/huggingface/transformers/issues/46188)] - Update self-comment-ci ([#​46137](https://redirect.github.com/huggingface/transformers/issues/46137)) by [@​guarin](https://redirect.github.com/guarin) in \[[#​46137](https://redirect.github.com/huggingface/transformers/issues/46137)] #### Significant community contributions The following contributors have made significant changes to the library over the last release: - [@​filipinescu](https://redirect.github.com/filipinescu) - \[docs] Romanian translation of `contributing.md`, `modular_transformers.md`, `multimodal_processing.md`, `add_vision_processing_components.md`, `add_audio_processing_components.md`, `modeling_rules.md`, `model_output_tracing.md`, `auto_docstring.md`, `testing.md`, `pr_checks.md` and `add_new_model.md` . ([#​46345](https://redirect.github.com/huggingface/transformers/issues/46345)) - \[docs] Romanian translation of `weightconverter.md`, `models.md`, `custom_models.md`, `monkey_patching.md`, `fusion_mapping.md`, `how_to_hack_models.md`, `model_sharing.md` and `serialization.md`. ([#​46309](https://redirect.github.com/huggingface/transformers/issues/46309)) - Romanian translation of README.md, index.md, installation.md, \_config.py and quicktour.md. ([#​46166](https://redirect.github.com/huggingface/transformers/issues/46166)) - [@​remi-or](https://redirect.github.com/remi-or) - \[CB] \[Major] Rework manager to have clearer control flow + handle TP ([#​46070](https://redirect.github.com/huggingface/transformers/issues/46070)) - [@​thisisiron](https://redirect.github.com/thisisiron) - Add Deepseek-OCR-2 model ([#​45075](https://redirect.github.com/huggingface/transformers/issues/45075)) - [@​kaixuanliu](https://redirect.github.com/kaixuanliu) - Add Expectations for pipeline token classification tests ([#​46151](https://redirect.github.com/huggingface/transformers/issues/46151)) - fix series of bugs for model parallel beam search ([#​46280](https://redirect.github.com/huggingface/transformers/issues/46280)) - add more generic support for distributed trainer tests ([#​46109](https://redirect.github.com/huggingface/transformers/issues/46109)) - add XPU Expectations for florence2 and lfm2\_vl model test ([#​46275](https://redirect.github.com/huggingface/transformers/issues/46275)) - Fix model parallel issue for altclip model and ChineseClip model ([#​45487](https://redirect.github.com/huggingface/transformers/issues/45487)) - Model parallel fix ([#​46230](https://redirect.github.com/huggingface/transformers/issues/46230)) - fix(hrm\_text): Add XPU Expectations for tests ([#​46214](https://redirect.github.com/huggingface/transformers/issues/46214)) - Fix model parallel bugs for Gemma4 ([#​45817](https://redirect.github.com/huggingface/transformers/issues/45817)) - Fix bnb 4bit/8bit quantization drop chunked tensors bug ([#​46210](https://redirect.github.com/huggingface/transformers/issues/46210)) - fix model parallel device mismatch issue in `create_bidirectional_mask` ([#​46221](https://redirect.github.com/huggingface/transformers/issues/46221)) - Fix a regression in encoder-decoder generation cache initialization ([#​46111](https://redirect.github.com/huggingface/transformers/issues/46111)) - [@​shadeMe](https://redirect.github.com/shadeMe) - feat: Add support for JetBrains' `Mellum` v2 code generation model ([#​46112](https://redirect.github.com/huggingface/transformers/issues/46112)) - [@​vasqu](https://redirect.github.com/vasqu) - \[`Revert`] FSDP+Dtensor refactor related changes ([#​46246](https://redirect.github.com/huggingface/transformers/issues/46246)) - \[`Configs`] Fix layer type validation to include its mlp counterpart ([#​46220](https://redirect.github.com/huggingface/transformers/issues/46220)) - [@​zRzRzRzRzRzRzR](https://redirect.github.com/zRzRzRzRzRzRzR) - \[GLM-4.6V] Update with GLM-GA Processor ([#​46184](https://redirect.github.com/huggingface/transformers/issues/46184)) - [@​eustlb](https://redirect.github.com/eustlb) - 🚨 \[ALM] Add base model without head ([#​45534](https://redirect.github.com/huggingface/transformers/issues/45534)) ### [`v5.10.0`](https://redirect.github.com/huggingface/transformers/compare/v5.9.0...v5.10.0) [Compare Source](https://redirect.github.com/huggingface/transformers/compare/v5.9.0...v5.10.0) ### [`v5.9.0`](https://redirect.github.com/huggingface/transformers/releases/tag/v5.9.0) [Compare Source](https://redirect.github.com/huggingface/transformers/compare/v5.8.1...v5.9.0) ### 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](https://huggingface.co/docs/transformers/main/en/model_doc/cohere2_moe) - Add new cohere2\_moe model ([#​46115](https://redirect.github.com/huggingface/transformers/issues/46115)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in [#​46115](https://redirect.github.com/huggingface/transformers/pull/46115) ##### Parakeet tdt ([#​44171](https://redirect.github.com/huggingface/transformers/issues/44171)) - Parakeet tdt ([#​44171](https://redirect.github.com/huggingface/transformers/issues/44171)) by [@​lmaksym](https://redirect.github.com/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](https://huggingface.co/docs/transformers/main/en/model_doc/hrm_text) | [Paper](https://huggingface.co/papers/2506.21734) - Add hrm text ([#​46025](https://redirect.github.com/huggingface/transformers/issues/46025)) by [@​abcd1927](https://redirect.github.com/abcd1927) in [#​46025](https://redirect.github.com/huggingface/transformers/pull/46025) #### Breaking changes The `text_embeds` input for SAM3, EdgeTAM, and SAM3-Lite-Text models now expects full text embeddings instead of just pooler outputs, aligning with other models in the library — users must update their inputs accordingly. - 🚨Fix memory leaks caused by lru decorators in vision models ([#​45922](https://redirect.github.com/huggingface/transformers/issues/45922)) by [@​yonigozlan](https://redirect.github.com/yonigozlan) #### Audio Audio support was expanded with the addition of AudioFlamingoNext model checkpoints and improved compilability of audio/vision encoders via standalone pure functions. Additional improvements include better error messaging when loading audio from video files and new documentation for audio/video processors. - user friendly error when loading audio from video ([#​45221](https://redirect.github.com/huggingface/transformers/issues/45221)) by [@​eustlb](https://redirect.github.com/eustlb) in \[[#​45221](https://redirect.github.com/huggingface/transformers/issues/45221)] - \[docs] adding audio/video processors ([#​45795](https://redirect.github.com/huggingface/transformers/issues/45795)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​45795](https://redirect.github.com/huggingface/transformers/issues/45795)] - Support Audio Flamingo Next checkpoints ([#​44830](https://redirect.github.com/huggingface/transformers/issues/44830)) by [@​lashahub](https://redirect.github.com/lashahub) in \[[#​44830](https://redirect.github.com/huggingface/transformers/issues/44830)] - Extract dynamic vision/audio tensors into standalone pure functions ([#​45396](https://redirect.github.com/huggingface/transformers/issues/45396)) by [@​IlyasMoutawwakil](https://redirect.github.com/IlyasMoutawwakil) in \[[#​45396](https://redirect.github.com/huggingface/transformers/issues/45396)] #### Generation Fixed generation issues including `inputs_embeds` and `per_layer_inputs` handling for Gemma4, an `AttributeError` in RAG's `generate()` caused by missing config fields, and flaky VLM generation tests by blocking special image tokens during sampling. - Fix Gemma4 generation from inputs\_embeds and per\_layer\_inputs ([#​46049](https://redirect.github.com/huggingface/transformers/issues/46049)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46049](https://redirect.github.com/huggingface/transformers/issues/46049)] - Fix AttributeError in RAG generate() for missing config fields ([#​46035](https://redirect.github.com/huggingface/transformers/issues/46035)) by [@​Sriniketh24](https://redirect.github.com/Sriniketh24) in \[[#​46035](https://redirect.github.com/huggingface/transformers/issues/46035)] - Block image\_start/end\_token\_id in generation test sampling ([#​45914](https://redirect.github.com/huggingface/transformers/issues/45914)) by [@​Rocketknight1](https://redirect.github.com/Rocketknight1) in \[[#​45914](https://redirect.github.com/huggingface/transformers/issues/45914)] #### Bugfixes and improvements - Remove mask visualization tool from `masking_utils.py` ([#​46066](https://redirect.github.com/huggingface/transformers/issues/46066)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​46066](https://redirect.github.com/huggingface/transformers/issues/46066)] - fix: owned\_by field in GET /v1/models returns list instead of string ([#​46006](https://redirect.github.com/huggingface/transformers/issues/46006)) by [@​nileshpatil6](https://redirect.github.com/nileshpatil6) in \[[#​46006](https://redirect.github.com/huggingface/transformers/issues/46006)] - \[CB] Remove OpenTelemetry ([#​45984](https://redirect.github.com/huggingface/transformers/issues/45984)) by [@​remi-or](https://redirect.github.com/remi-or) in \[[#​45984](https://redirect.github.com/huggingface/transformers/issues/45984)] - docs(readme): use canonical `huggingface.co` domain in prose links ([#​46042](https://redirect.github.com/huggingface/transformers/issues/46042)) by [@​kiwigitops](https://redirect.github.com/kiwigitops) in \[[#​46042](https://redirect.github.com/huggingface/transformers/issues/46042)] - Fix remaining RAG doc examples that crash on current transformers ([#​46044](https://redirect.github.com/huggingface/transformers/issues/46044)) by [@​Sriniketh24](https://redirect.github.com/Sriniketh24) in \[[#​46044](https://redirect.github.com/huggingface/transformers/issues/46044)] - Init the actual tensor, not a copy ([#​46030](https://redirect.github.com/huggingface/transformers/issues/46030)) by [@​Rocketknight1](https://redirect.github.com/Rocketknight1) in \[[#​46030](https://redirect.github.com/huggingface/transformers/issues/46030)] - docs: sync legacy ACL anthology URLs and update metrics across i18n READMEs ([#​46027](https://redirect.github.com/huggingface/transformers/issues/46027)) by [@​irfaan101](https://redirect.github.com/irfaan101) in \[[#​46027](https://redirect.github.com/huggingface/transformers/issues/46027)] - \[MultimodalLM] add language\_model to the get/set\_input\_embeddings logic ([#​46029](https://redirect.github.com/huggingface/transformers/issues/46029)) by [@​eustlb](https://redirect.github.com/eustlb) in \[[#​46029](https://redirect.github.com/huggingface/transformers/issues/46029)] - \[`HRM Text`] Add integration tests ([#​46033](https://redirect.github.com/huggingface/transformers/issues/46033)) by [@​vasqu](https://redirect.github.com/vasqu) in \[[#​46033](https://redirect.github.com/huggingface/transformers/issues/46033)] - hy\_v3: add XPU expectations ([#​45858](https://redirect.github.com/huggingface/transformers/issues/45858)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​45858](https://redirect.github.com/huggingface/transformers/issues/45858)] - exaone4\_5: add XPU expectations ([#​45890](https://redirect.github.com/huggingface/transformers/issues/45890)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​45890](https://redirect.github.com/huggingface/transformers/issues/45890)] - hyperclovax: add XPU Expectations for CI test ([#​45926](https://redirect.github.com/huggingface/transformers/issues/45926)) by [@​kaixuanliu](https://redirect.github.com/kaixuanliu) in \[[#​45926](https://redirect.github.com/huggingface/transformers/issues/45926)] - chore(ci): remove dead env vars from circleci-failure-summary-comment.yml ([#​45972](https://redirect.github.com/huggingface/transformers/issues/45972)) by [@​XciD](https://redirect.github.com/XciD) in \[[#​45972](https://redirect.github.com/huggingface/transformers/issues/45972)] - \[CB] \[Major] Add tensor paralellism ([#​45821](https://redirect.github.com/huggingface/transformers/issues/45821)) by [@​remi-or](https://redirect.github.com/remi-or) in \[[#​45821](https://redirect.github.com/huggingface/transformers/issues/45821)] - docs: update models architecture count and sync ACL anthology URLs ([#​46001](https://redirect.github.com/huggingface/transformers/issues/46001)) by [@​irfaan101](https://redirect.github.com/irfaan101) in \[[#​46001](https://redirect.github.com/huggingface/transformers/issues/46001)] - bugfix(ci): avoid E2BIG in pr\_slow\_ci\_suggestion ([#​45983](https://redirect.github.com/huggingface/transformers/issues/45983)) by [@​tarekziade](https://redirect.github.com/tarekziade) in \[[#​45983](https://redirect.github.com/huggingface/transformers/issues/45983)] - RFDetr - use correct Roboflow org for release ([#​45946](https://redirect.github.com/huggingface/transformers/issues/45946)) by [@​sbucaille](https://redirect.github.com/sbucaille) in \[[#​45946](https://redirect.github.com/huggingface/transformers/issues/45946)] - docs: Fix formatting issues in weightconverter.md ([#​45988](https://redirect.github.com/huggingface/transformers/issues/45988)) by [@​ArjunSrivastava1](https://redirect.github.com/ArjunSrivastava1) in \[[#​45988](https://redirect.github.com/huggingface/transformers/issues/45988)] - Fix colqwen2 test ([#​45981](https://redirect.github.com/huggingface/transformers/issues/45981)) by [@​IlyasMoutawwakil](https://redirect.github.com/IlyasMoutawwakil) in \[[#​45981](https://redirect.github.com/huggingface/transformers/issues/45981)] - Fix M-RoPE device mismatch in Qwen3VL family under FSDP2 CPU offload ([#​45861](https://redirect.github.com/huggingface/transformers/issues/45861)) by [@​jamesbraza](https://redirect.github.com/jamesbraza) in \[[#​45861](https://redirect.github.com/huggingface/transformers/issues/45861)] - \[docs] chat template prefill ([#​45947](https://redirect.github.com/huggingface/transformers/issues/45947)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​45947](https://redirect.github.com/huggingface/transformers/issues/45947)] - \[docs] decode fast path ([#​45899](https://redirect.github.com/huggingface/transformers/issues/45899)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​45899](https://redirect.github.com/huggingface/transformers/issues/45899)] - fix: restore `_attn_implementation `and fix request offset in `generate_batch()` ([#​45943](https://redirect.github.com/huggingface/transformers/issues/45943)) by [@​sergiopaniego](https://redirect.github.com/sergiopaniego) in \[[#​45943](https://redirect.github.com/huggingface/transformers/issues/45943)] - Expose `per_layer_inputs` for every Gemma4 variants ([#​45927](https://redirect.github.com/huggingface/transformers/issues/45927)) by [@​Cyrilvallez](https://redirect.github.com/Cyrilvallez) in \[[#​45927](https://redirect.github.com/huggingface/transformers/issues/45927)] - chore: update benchmark\_v2.yml ([#​45966](https://redirect.github.com/huggingface/transformers/issues/45966)) by [@​hf-security-analysis](https://redirect.github.com/hf-security-analysis)\[bot] in \[[#​45966](https://redirect.github.com/huggingface/transformers/issues/45966)] - fix(ci): set persist-credentials: false on actions/checkout and close remaining template injection findings ([#​45964](https://redirect.github.com/huggingface/transformers/issues/45964)) by [@​XciD](https://redirect.github.com/XciD) in \[[#​45964](https://redirect.github.com/huggingface/transformers/issues/45964)] - chore(ci): set default workflow permissions to contents: read ([#​45961](https://redirect.github.com/huggingface/transformers/issues/45961)) by [@​XciD](https://redirect.github.com/XciD) in \[[#​45961](https://redirect.github.com/huggingface/transformers/issues/45961)] - fix(ci): remove template injection on pull\_request\_target workflows ([#​45956](https://redirect.github.com/huggingface/transformers/issues/45956)) by [@​XciD](https://redirect.github.com/XciD) in \[[#​45956](https://redirect.github.com/huggingface/transformers/issues/45956)] - chore(ci): pin all GitHub Actions and reusable workflows by SHA ([#​45955](https://redirect.github.com/huggingface/transformers/issues/45955)) by [@​XciD](https://redirect.github.com/XciD) in \[[#​45955](https://redirect.github.com/huggingface/transformers/issues/45955)] - \[docs] ALMModelTest ([#​45900](https://redirect.github.com/huggingface/transformers/issues/45900)) by [@​stevhliu](https://redirect.github.com/stevhliu) in \[[#​45900](https://redirect.github.com/huggingface/transformers/issues/45900)] - Enhance apply\_chat\_template to support custom field prefilling (reasoning\_content, thinking, etc.) ([#​45896](https://redirect.github.com/huggingface/transformers/issues/45896)) by [@​Mamiglia](https://redirect.github.com/Mamiglia) in \[[#​45896](https://redirect.github.com/huggingface/transformers/issues/45896)] - BUGFIX: Support hubert models that don't have conv\_pos\_batch\_norm configured ([#​45921](https://redirect.github.com/huggingface/transformers/issues/45921)) by [@​igordertigor](https://redirect.github.com/igordertigor) in \[[#​45921](https://redirect.github.com/huggingface/transformers/issues/45921)] - Revert 45777 ([#​45942](https://redirect.github.com/huggingface/transformers/issues/45942)) by [@​Rocketknight1](https://redirect.github.com/Rocketknight1) in \[[#​45942](https://redirect.github.com/huggingface/transformers/issues/45942)] - pass the otel secrets ([#​45933](https://redirect.github.com/huggingface/transformers/issues/45933)) by [@​tarekziade](https://redirect.github.com/tarekziade) in \[[#​45933](https://redirect.github.com/huggingface/transformers/issues/45933)] - Add initial torch\_tpu backend support ([#​45918](https://redirect.github.com/huggingface/transformers/issues/45918)) by [@​tengomucho](https://redirect.github.com/tengomucho) in \[[#​45918](https://redirect.github.com/huggingface/transformers/issues/45918)] - \[CB] Hide activation footprint by using the CUDA graph pool ([#​45911](https://redirect.github.com/huggingface/transformers/issues/45911)) by [@​remi-or](https://redirect.github.com/remi-or) in \[[#​45911](https://redirect.github.com/huggingface/transformers/issues/45911)] - Require input\_ids for repetition penalty ([#​45389](https://redirect.github.com/huggingface/transformers/issues/45389)) by [@​ruben-aghayan](https://redirect.github.com/ruben-aghayan) in \[[#​45389](https://redirect.github.com/huggingface/t > ✂ **Note** > > PR body was truncated to here. </details> --- ### Configuration 📅 **Schedule**: (in timezone Etc/UTC) - Branch creation - At any time (no schedule defined) - Automerge - At any time (no schedule defined) 🚦 **Automerge**: Disabled by config. Please merge this manually once you are satisfied. ♻ **Rebasing**: Whenever PR becomes conflicted, or you tick the rebase/retry checkbox. 🔕 **Ignore**: Close this PR and you won't be reminded about this update again. --- - [ ] <!-- rebase-check -->If you want to rebase/retry this PR, check this box --- This PR was generated by [Mend Renovate](https://mend.io/renovate/). View the [repository job log](https://developer.mend.io/github/apache/texera). <!--renovate-debug:eyJjcmVhdGVkSW5WZXIiOiI0NC40OS4wIiwidXBkYXRlZEluVmVyIjoiNDQuNDkuMCIsInRhcmdldEJyYW5jaCI6Im1haW4iLCJsYWJlbHMiOlsiZGVwZW5kZW5jaWVzIiwicmVsZWFzZS92MS4yIiwic2VjdXJpdHkiXX0=--> -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
