元記事の説明文
<h1>Release v5.10.1</h1>
<p>v5.10.0 was yanked as we publish on a corrupted branch. Sorry everyone, this happens when we rush a release!!!</p>
<h2>New Model additions</h2>
<h3>Gemma4 unified+ Gemma4 MTP</h3>
<a href="https://private-user-images.githubusercontent.com/48595927/602317672-5e3ee940-f78d-4343-ac7a-889930800aa6.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.FVe8xb9pZoOx6ZIQ7mXPhmiTqwHyLjwPA9vNDi9qR4o" rel="noopener noreferrer" target="_blank"><img alt="image" height="400" src="https://private-user-images.githubusercontent.com/48595927/602317672-5e3ee940-f78d-4343-ac7a-889930800aa6.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.FVe8xb9pZoOx6ZIQ7mXPhmiTqwHyLjwPA9vNDi9qR4o" style="height: auto;" width="2000" /></a>
<p>Gemma 4 12B Unified is an <strong>encoder-free</strong> multimodal model with pretrained and instruction-tuned variants. Unlike <a href="https://github.com/huggingface/transformers/blob/v5.10.1/gemma4">standard Gemma 4</a>, 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.</p>
<p>Key differences from standard Gemma 4:</p>
<ul>
<li><strong>No Vision Tower</strong>: Raw pixel patches are projected directly into LM space via a <code>Dense + LayerNorm</code> pipeline with factorized 2D positional embeddings, replacing the vision encoder.</li>
<li><strong>No Audio Tower</strong>: Raw 16 kHz waveform samples are chunked into fixed-length frames and projected through a simple <code>RMSNorm → Linear</code> pipeline, replacing the mel spectrogram + Conformer encoder.</li>
<li><strong>Shared Multimodal Pipeline</strong>: Both vision and audio use the same <code>Gemma4UnifiedMultimodalEmbedder</code> (RMSNorm → Linear) for the final projection to text hidden space.</li>
</ul>
<p>You can find the original Gemma 4 12B Unified checkpoints under the <a href="https://huggingface.co/collections/google/gemma-4" rel="nofollow">Gemma 4</a> release.</p>
<ul>
<li>who needs encoders? (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46385">#46385</a>) by <a class="user-mention notranslate" href="https://github.com/douglas-reid">@douglas-reid</a> <a class="user-mention notranslate" href="https://github.com/sgerrard">@sgerrard</a> <a class="user-mention notranslate" href="https://github.com/vasqu">@vasqu</a> <a class="user-mention notranslate" href="https://github.com/molbap">@molbap</a></li>
</ul>
<h3>Sapiens2</h3>
<p>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.</p>
<p><strong>Links:</strong> <a href="https://huggingface.co/docs/transformers/main/en/model_doc/sapiens2" rel="nofollow">Documentation</a> | <a href="https://huggingface.co/papers/2604.21681" rel="nofollow">Paper</a></p>
<ul>
<li>Add Sapiens2 Model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45919">#45919</a>) by <a class="user-mention notranslate" href="https://github.com/guarin">@guarin</a> in <a href="https://github.com/huggingface/transformers/pull/45919">#45919</a></li>
</ul>
<h3>DeepSeek-OCR-2</h3>
<p>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.</p>
<p><strong>Links:</strong> <a href="https://huggingface.co/docs/transformers/main/en/model_doc/deepseek_ocr2" rel="nofollow">Documentation</a></p>
<ul>
<li>Add Deepseek-OCR-2 model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45075">#45075</a>) by <a class="user-mention notranslate" href="https://github.com/thisisiron">@thisisiron</a> in <a href="https://github.com/huggingface/transformers/pull/45075">#45075</a></li>
</ul>
<h3>Mellum</h3>
<p>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.</p>
<p><strong>Links:</strong> <a href="https://huggingface.co/docs/transformers/main/en/model_doc/mellum" rel="nofollow">Documentation</a></p>
<ul>
<li>feat: Add support for JetBrains' <code>Mellum</code> v2 code generation model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46112">#46112</a>) by <a class="user-mention notranslate" href="https://github.com/shadeMe">@shadeMe</a> in <a href="https://github.com/huggingface/transformers/pull/46112">#46112</a></li>
</ul>
<h2>Breaking changes</h2>
<p>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.</p>
<ul>
<li>🚨 Fix float16 overflow in Gemma4 vision pooler (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46277">#46277</a>) by <a class="user-mention notranslate" href="https://github.com/Bluear7878">@Bluear7878</a></li>
</ul>
<p>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.</p>
<ul>
<li>🚨 [ALM] Add base model without head (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45534">#45534</a>) by <a class="user-mention notranslate" href="https://github.com/eustlb">@eustlb</a></li>
</ul>
<h2>Parallelization</h2>
<p>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 <code>from_pretrained</code> was introduced.</p>
<ul>
<li>Fix dsv4 dequant + tp/ep (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46378">#46378</a>) by <a class="user-mention notranslate" href="https://github.com/IlyasMoutawwakil">@IlyasMoutawwakil</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46378">#46378</a>]</li>
<li>[CB] [Major] Rework manager to have clearer control flow + handle TP (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46070">#46070</a>) by <a class="user-mention notranslate" href="https://github.com/remi-or">@remi-or</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46070">#46070</a>]</li>
<li>fix series of bugs for model parallel beam search (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46280">#46280</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46280">#46280</a>]</li>
<li>Fix model parallel issue for altclip model and ChineseClip model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45487">#45487</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45487">#45487</a>]</li>
<li>Model parallel fix (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46230">#46230</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46230">#46230</a>]</li>
<li>[<code>Revert</code>] FSDP+Dtensor refactor related changes (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46246">#46246</a>) by <a class="user-mention notranslate" href="https://github.com/vasqu">@vasqu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46246">#46246</a>]</li>
<li>Fix model parallel bugs for Gemma4 (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45817">#45817</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45817">#45817</a>]</li>
<li>init FSDP through from_pretrained (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46102">#46102</a>) by <a class="user-mention notranslate" href="https://github.com/3outeille">@3outeille</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46102">#46102</a>]</li>
<li>fix model parallel device mismatch issue in <code>create_bidirectional_mask</code> (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46221">#46221</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46221">#46221</a>]</li>
<li>Trainer.compute_loss: fix loss over-counting under TP and EP-as-TP (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45994">#45994</a>) by <a class="user-mention notranslate" href="https://github.com/AmineDiro">@AmineDiro</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45994">#45994</a>]</li>
<li>Fix caching allocator warmup byte estimation for EP model loading (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46149">#46149</a>) by <a class="user-mention notranslate" href="https://github.com/sywangyi">@sywangyi</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46149">#46149</a>]</li>
</ul>
<h2>Cache</h2>
<p>Fixed a regression in encoder-decoder cache initialization where the decoder config was incorrectly applied to the cross-attention cache, and resolved a <code>RuntimeError</code> 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.</p>
<ul>
<li>fix: cache warmup <code>RuntimeError</code> on mps (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46239">#46239</a>) by <a class="user-mention notranslate" href="https://github.com/McPatate">@McPatate</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46239">#46239</a>]</li>
<li>Make more tests work with read-only cache (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46299">#46299</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46299">#46299</a>]</li>
<li>Update a test to avoid writing to the default xet cache (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46250">#46250</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46250">#46250</a>]</li>
<li>Fix a regression in encoder-decoder generation cache initialization (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46111">#46111</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46111">#46111</a>]</li>
</ul>
<h2>Quantization</h2>
<p>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.</p>
<ul>
<li>DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45634">#45634</a>) by <a class="user-mention notranslate" href="https://github.com/IlyasMoutawwakil">@IlyasMoutawwakil</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45634">#45634</a>]</li>
<li>Fix fp8 moe reverse substring (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46265">#46265</a>) by <a class="user-mention notranslate" href="https://github.com/ArthurZucker">@ArthurZucker</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46265">#46265</a>]</li>
<li>Fix bnb 4bit/8bit quantization drop chunked tensors bug (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46210">#46210</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46210">#46210</a>]</li>
</ul>
<h2>Bugfixes and improvements</h2>
<ul>
<li>Fix wrong changes produced by style/repo. check bot (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46371">#46371</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46371">#46371</a>]</li>
<li>Fix path traversal when saving Bark voice preset embeddings (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46237">#46237</a>) by <a class="user-mention notranslate" href="https://github.com/LinZiyuu">@LinZiyuu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46237">#46237</a>]</li>
<li>Pass library_name/version to Hub calls via a shared HfApi (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46318">#46318</a>) by <a class="user-mention notranslate" href="https://github.com/Wauplin">@Wauplin</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46318">#46318</a>]</li>
<li>docs: update ACL Anthology URL in CITATION.cff (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46352">#46352</a>) by <a class="user-mention notranslate" href="https://github.com/irfaan101">@irfaan101</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46352">#46352</a>]</li>
<li>[docs] contributing (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45465">#45465</a>) by <a class="user-mention notranslate" href="https://github.com/stevhliu">@stevhliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45465">#45465</a>]</li>
<li>[docs] Romanian translation of <code>contributing.md</code>, <code>modular_transformers.md</code>, <code>multimodal_processing.md</code>, <code>add_vision_processing_components.md</code>, <code>add_audio_processing_components.md</code>, <code>modeling_rules.md</code>, <code>model_output_tracing.md</code>, <code>auto_docstring.md</code>, <code>testing.md</code>, <code>pr_checks.md</code> and <code>add_new_model.md</code> . (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46345">#46345</a>) by <a class="user-mention notranslate" href="https://github.com/filipinescu">@filipinescu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46345">#46345</a>]</li>
<li>[docs] xpu continuous batching (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46334">#46334</a>) by <a class="user-mention notranslate" href="https://github.com/stevhliu">@stevhliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46334">#46334</a>]</li>
<li>Fix incorrect attribute mapping relationships in GLM MoE DSA Config (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46338">#46338</a>) by <a class="user-mention notranslate" href="https://github.com/Dovis01">@Dovis01</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46338">#46338</a>]</li>
<li>Fix grammar typos in Whisper documentation (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46336">#46336</a>) by <a class="user-mention notranslate" href="https://github.com/calliec-1223">@calliec-1223</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46336">#46336</a>]</li>
<li>[docs] update num_items_in_batch for causal LMs (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46335">#46335</a>) by <a class="user-mention notranslate" href="https://github.com/stevhliu">@stevhliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46335">#46335</a>]</li>
<li>Update compressed tensors minimum version (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46342">#46342</a>) by <a class="user-mention notranslate" href="https://github.com/SunMarc">@SunMarc</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46342">#46342</a>]</li>
<li>Fix _is_package_available reporting available without a version (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46125">#46125</a>) by <a class="user-mention notranslate" href="https://github.com/blipbyte">@blipbyte</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46125">#46125</a>]</li>
<li>remove sec (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46346">#46346</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46346">#46346</a>]</li>
<li>fix: include transitive relative imports when loading from local directory (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46022">#46022</a>) by <a class="user-mention notranslate" href="https://github.com/trducng">@trducng</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46022">#46022</a>]</li>
<li>perf(feature_extraction_sequence): skip re-splitting already-batched numpy arrays in pad() (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46329">#46329</a>) by <a class="user-mention notranslate" href="https://github.com/Anai-Guo">@Anai-Guo</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46329">#46329</a>]</li>
<li>[Zamba] Support attn_implementation dispatch (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46317">#46317</a>) by <a class="user-mention notranslate" href="https://github.com/YangKai0616">@YangKai0616</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46317">#46317</a>]</li>
<li>Fix TestAppRoutes test failures caused by deprecated asyncio.get_event_loop() on Python 3.10+ (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46340">#46340</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46340">#46340</a>]</li>
<li>[Qwen3VL] Fix video token placeholder: use self.video_token instead of hardcoded "<|placeholder|>" (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46296">#46296</a>) by <a class="user-mention notranslate" href="https://github.com/kpal002">@kpal002</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46296">#46296</a>]</li>
<li>chore(linter): fixes for rule 16 (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46023">#46023</a>) by <a class="user-mention notranslate" href="https://github.com/tarekziade">@tarekziade</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46023">#46023</a>]</li>
<li>[docs] Romanian translation of <code>weightconverter.md</code>, <code>models.md</code>, <code>custom_models.md</code>, <code>monkey_patching.md</code>, <code>fusion_mapping.md</code>, <code>how_to_hack_models.md</code>, <code>model_sharing.md</code> and <code>serialization.md</code>. (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46309">#46309</a>) by <a class="user-mention notranslate" href="https://github.com/filipinescu">@filipinescu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46309">#46309</a>]</li>
<li>Normalize CUDA OOM errors when comparing commit failures in check_bad_commit (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46322">#46322</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46322">#46322</a>]</li>
<li>Fix unhandled exception noise from background safetensors conversion thread (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45752">#45752</a>) by <a class="user-mention notranslate" href="https://github.com/dhruv7477">@dhruv7477</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45752">#45752</a>]</li>
<li>Add Expectations for pipeline token classification tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46151">#46151</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46151">#46151</a>]</li>
<li>[docs] fix auto-add release dates (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46283">#46283</a>) by <a class="user-mention notranslate" href="https://github.com/zucchini-nlp">@zucchini-nlp</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46283">#46283</a>]</li>
<li>Separate pip command syntax for notebook and CLI tabs in Quickstart (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46243">#46243</a>) by <a class="user-mention notranslate" href="https://github.com/pvelayudhan">@pvelayudhan</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46243">#46243</a>]</li>
<li>Romanian translation of README.md, index.md, installation.md, _config.py and quicktour.md. (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46166">#46166</a>) by <a class="user-mention notranslate" href="https://github.com/filipinescu">@filipinescu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46166">#46166</a>]</li>
<li>Fall back to flat kwarg when modality dict is passed without it (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46195">#46195</a>) by <a class="user-mention notranslate" href="https://github.com/Ace3Z">@Ace3Z</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46195">#46195</a>]</li>
<li>Fix load_adapter OOM caused by full-model warmup sizing (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46145">#46145</a>) by <a class="user-mention notranslate" href="https://github.com/Yooniel">@Yooniel</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46145">#46145</a>]</li>
<li>Replace assert with raise ImportError for optuna/ray dependency checks (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46263">#46263</a>) by <a class="user-mention notranslate" href="https://github.com/SebTardif">@SebTardif</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46263">#46263</a>]</li>
<li>chore(linter): respect TRF017 modeling rule (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46260">#46260</a>) by <a class="user-mention notranslate" href="https://github.com/tarekziade">@tarekziade</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46260">#46260</a>]</li>
<li>Delete dead code in qwen-vl series (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45827">#45827</a>) by <a class="user-mention notranslate" href="https://github.com/zucchini-nlp">@zucchini-nlp</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45827">#45827</a>]</li>
<li>qa: fix ty caching and align CI with local run (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46278">#46278</a>) by <a class="user-mention notranslate" href="https://github.com/tarekziade">@tarekziade</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46278">#46278</a>]</li>
<li>Guard DeviceMesh import in continuous batching (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46205">#46205</a>) by <a class="user-mention notranslate" href="https://github.com/danyalahmed1995">@danyalahmed1995</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46205">#46205</a>]</li>
<li>Processor compatibility with vLLM (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46258">#46258</a>) by <a class="user-mention notranslate" href="https://github.com/zucchini-nlp">@zucchini-nlp</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46258">#46258</a>]</li>
<li>Fix PR CI workflow cancellation condition (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46276">#46276</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46276">#46276</a>]</li>
<li>[fix] toctree (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46106">#46106</a>) by <a class="user-mention notranslate" href="https://github.com/stevhliu">@stevhliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46106">#46106</a>]</li>
<li>add more generic support for distributed trainer tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46109">#46109</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46109">#46109</a>]</li>
<li>add XPU Expectations for florence2 and lfm2_vl model test (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46275">#46275</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46275">#46275</a>]</li>
<li>Fix <code>StaticCache</code> building an empty layer list when <code>num_kv_shared_layers == 0</code> (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46235">#46235</a>) by <a class="user-mention notranslate" href="https://github.com/tengomucho">@tengomucho</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46235">#46235</a>]</li>
<li>Fix inverted assertion in remove_handler (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46227">#46227</a>) by <a class="user-mention notranslate" href="https://github.com/SebTardif">@SebTardif</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46227">#46227</a>]</li>
<li>[ShieldGemma2] Support attn_implementation dispatch (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46069">#46069</a>) by <a class="user-mention notranslate" href="https://github.com/YangKai0616">@YangKai0616</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46069">#46069</a>]</li>
<li>[Gemma4] Replace one-hot matmul with F.embedding in position embeddings (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46176">#46176</a>) by <a class="user-mention notranslate" href="https://github.com/Sriniketh24">@Sriniketh24</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46176">#46176</a>]</li>
<li>fix: kosmos2.5: properly expand embeddings table (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45835">#45835</a>) by <a class="user-mention notranslate" href="https://github.com/nunq">@nunq</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45835">#45835</a>]</li>
<li>find pytest launch error in torch 2.13.0.dev20260526 (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46252">#46252</a>) by <a class="user-mention notranslate" href="https://github.com/sywangyi">@sywangyi</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46252">#46252</a>]</li>
<li>[Test][Kosmos2.5] Add XPU expectations for integration tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46135">#46135</a>) by <a class="user-mention notranslate" href="https://github.com/YangKai0616">@YangKai0616</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46135">#46135</a>]</li>
<li>Support FA2 flash_attn_with_kvcache for XPU continuous batching (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46028">#46028</a>) by <a class="user-mention notranslate" href="https://github.com/YangKai0616">@YangKai0616</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46028">#46028</a>]</li>
<li>[<code>Configs</code>] Fix layer type validation to include its mlp counterpart (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46220">#46220</a>) by <a class="user-mention notranslate" href="https://github.com/vasqu">@vasqu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46220">#46220</a>]</li>
<li>Fix <code>num_items_in_batch</code> over-counting for causal LM losses (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46204">#46204</a>) by <a class="user-mention notranslate" href="https://github.com/qgallouedec">@qgallouedec</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46204">#46204</a>]</li>
<li>RF-DETR doc fixes (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46244">#46244</a>) by <a class="user-mention notranslate" href="https://github.com/merveenoyan">@merveenoyan</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46244">#46244</a>]</li>
<li>Use <code>main</code> instead of commit SHA for now (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46241">#46241</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46241">#46241</a>]</li>
<li>Enable push event (to main) for PR CI workflow (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46240">#46240</a>) by <a class="user-mention notranslate" href="https://github.com/ydshieh">@ydshieh</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46240">#46240</a>]</li>
<li>fix(hrm_text): Add XPU Expectations for tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46214">#46214</a>) by <a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46214">#46214</a>]</li>
<li>[deepseek_v4] keep hc_head / sinks / position_bias in fp32 (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46198">#46198</a>) by <a class="user-mention notranslate" href="https://github.com/ArthurZucker">@ArthurZucker</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46198">#46198</a>]</li>
<li>Fix FSDP2 and distributed checkpointing imports for older PyTorch versions (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46141">#46141</a>) by <a class="user-mention notranslate" href="https://github.com/ryota-komatsu">@ryota-komatsu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46141">#46141</a>]</li>
<li>Fix Gemma4 Array Mask Indexing (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46203">#46203</a>) by <a class="user-mention notranslate" href="https://github.com/petecao">@petecao</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46203">#46203</a>]</li>
<li>utils: handle flash_attn missing from importlib packages_distributions without crashing (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45524">#45524</a>) by <a class="user-mention notranslate" href="https://github.com/SAY-5">@SAY-5</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45524">#45524</a>]</li>
<li>[AMD CI] revert AMD mi325 hf-workflows ref from SHA back to <a class="user-mention notranslate" href="https://github.com/main">@main</a> (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46213">#46213</a>) by <a class="user-mention notranslate" href="https://github.com/Abdennacer-Badaoui">@Abdennacer-Badaoui</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46213">#46213</a>]</li>
<li>[GLM-4.6V] Update with GLM-GA Processor (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46184">#46184</a>) by <a class="user-mention notranslate" href="https://github.com/zRzRzRzRzRzRzR">@zRzRzRzRzRzRzR</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46184">#46184</a>]</li>
<li>update xpu expectation for falcon mamba (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46086">#46086</a>) by <a class="user-mention notranslate" href="https://github.com/sywangyi">@sywangyi</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46086">#46086</a>]</li>
<li>chore: enable Dependabot weekly GitHub Actions bumps (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46157">#46157</a>) by @hf-dependantbot-rollout[bot] in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46157">#46157</a>]</li>
<li>Fix Gemma4 use_bidirectional_attention="all" mask behavior (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46079">#46079</a>) by <a class="user-mention notranslate" href="https://github.com/oliverholworthy">@oliverholworthy</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46079">#46079</a>]</li>
<li>Fix loading with only 1 device or distributed config (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46197">#46197</a>) by <a class="user-mention notranslate" href="https://github.com/Cyrilvallez">@Cyrilvallez</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46197">#46197</a>]</li>
<li>Fix TypeError on list-typed ignore_keys_at_rope_validation in RoPE config (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46142">#46142</a>) by <a class="user-mention notranslate" href="https://github.com/Charly21r">@Charly21r</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46142">#46142</a>]</li>
<li>Support XPU autocast dtype fallback for FlashAttention (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46199">#46199</a>) by <a class="user-mention notranslate" href="https://github.com/YangKai0616">@YangKai0616</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46199">#46199</a>]</li>
<li>Fix path traversal when saving named chat templates (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46191">#46191</a>) by <a class="user-mention notranslate" href="https://github.com/LinZiyuu">@LinZiyuu</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46191">#46191</a>]</li>
<li>Fix is_last off-by-one in MaskGenerationPipeline for partial batches (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46136">#46136</a>) by <a class="user-mention notranslate" href="https://github.com/J3r3myPerera">@J3r3myPerera</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46136">#46136</a>]</li>
<li>Fix wrong variable in check_model_type isinstance check (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46080">#46080</a>) by <a class="user-mention notranslate" href="https://github.com/SebTardif">@SebTardif</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46080">#46080</a>]</li>
<li>Enable passing kwargs through RoFormer models (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46171">#46171</a>) by <a class="user-mention notranslate" href="https://github.com/ir2718">@ir2718</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46171">#46171</a>]</li>
<li>Update cohere2_moe tp_plan (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46189">#46189</a>) by <a class="user-mention notranslate" href="https://github.com/Cyrilvallez">@Cyrilvallez</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46189">#46189</a>]</li>
<li>Update release tool (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46193">#46193</a>) by <a class="user-mention notranslate" href="https://github.com/Cyrilvallez">@Cyrilvallez</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46193">#46193</a>]</li>
<li>[loading] Fix base_model_prefix issues in conversions (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46067">#46067</a>) by <a class="user-mention notranslate" href="https://github.com/Cyrilvallez">@Cyrilvallez</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46067">#46067</a>]</li>
<li>Bump dev version (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46188">#46188</a>) by <a class="user-mention notranslate" href="https://github.com/Cyrilvallez">@Cyrilvallez</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46188">#46188</a>]</li>
<li>Update self-comment-ci (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46137">#46137</a>) by <a class="user-mention notranslate" href="https://github.com/guarin">@guarin</a> in [<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46137">#46137</a>]</li>
</ul>
<h2>Significant community contributions</h2>
<p>The following contributors have made significant changes to the library over the last release:</p>
<ul>
<li><a class="user-mention notranslate" href="https://github.com/filipinescu">@filipinescu</a>
<ul>
<li>[docs] Romanian translation of <code>contributing.md</code>, <code>modular_transformers.md</code>, <code>multimodal_processing.md</code>, <code>add_vision_processing_components.md</code>, <code>add_audio_processing_components.md</code>, <code>modeling_rules.md</code>, <code>model_output_tracing.md</code>, <code>auto_docstring.md</code>, <code>testing.md</code>, <code>pr_checks.md</code> and <code>add_new_model.md</code> . (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46345">#46345</a>)</li>
<li>[docs] Romanian translation of <code>weightconverter.md</code>, <code>models.md</code>, <code>custom_models.md</code>, <code>monkey_patching.md</code>, <code>fusion_mapping.md</code>, <code>how_to_hack_models.md</code>, <code>model_sharing.md</code> and <code>serialization.md</code>. (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46309">#46309</a>)</li>
<li>Romanian translation of README.md, index.md, installation.md, _config.py and quicktour.md. (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46166">#46166</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/remi-or">@remi-or</a>
<ul>
<li>[CB] [Major] Rework manager to have clearer control flow + handle TP (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46070">#46070</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/thisisiron">@thisisiron</a>
<ul>
<li>Add Deepseek-OCR-2 model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45075">#45075</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/kaixuanliu">@kaixuanliu</a>
<ul>
<li>Add Expectations for pipeline token classification tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46151">#46151</a>)</li>
<li>fix series of bugs for model parallel beam search (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46280">#46280</a>)</li>
<li>add more generic support for distributed trainer tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46109">#46109</a>)</li>
<li>add XPU Expectations for florence2 and lfm2_vl model test (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46275">#46275</a>)</li>
<li>Fix model parallel issue for altclip model and ChineseClip model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45487">#45487</a>)</li>
<li>Model parallel fix (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46230">#46230</a>)</li>
<li>fix(hrm_text): Add XPU Expectations for tests (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46214">#46214</a>)</li>
<li>Fix model parallel bugs for Gemma4 (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45817">#45817</a>)</li>
<li>Fix bnb 4bit/8bit quantization drop chunked tensors bug (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46210">#46210</a>)</li>
<li>fix model parallel device mismatch issue in <code>create_bidirectional_mask</code> (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46221">#46221</a>)</li>
<li>Fix a regression in encoder-decoder generation cache initialization (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46111">#46111</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/shadeMe">@shadeMe</a>
<ul>
<li>feat: Add support for JetBrains' <code>Mellum</code> v2 code generation model (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46112">#46112</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/vasqu">@vasqu</a>
<ul>
<li>[<code>Revert</code>] FSDP+Dtensor refactor related changes (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46246">#46246</a>)</li>
<li>[<code>Configs</code>] Fix layer type validation to include its mlp counterpart (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46220">#46220</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/zRzRzRzRzRzRzR">@zRzRzRzRzRzRzR</a>
<ul>
<li>[GLM-4.6V] Update with GLM-GA Processor (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/46184">#46184</a>)</li>
</ul>
</li>
<li><a class="user-mention notranslate" href="https://github.com/eustlb">@eustlb</a>
<ul>
<li>🚨 [ALM] Add base model without head (<a class="issue-link js-issue-link" href="https://github.com/huggingface/transformers/pull/45534">#45534</a>)</li>
</ul>
</li>
</ul>