Bump the pip group across 71 directories with 5 updates

[dependabot]

Bumps the pip group with 4 updates in the /.aitk/requirements directory: accelerate, cryptography, transformers and tornado.
Bumps the pip group with 1 update in the /Qwen-Qwen3.5-2B/cpu directory: transformers.
Bumps the pip group with 1 update in the /Qwen-Qwen3.5-2B/cuda directory: transformers.
Bumps the pip group with 1 update in the /Qwen-Qwen3.5-2B/webgpu directory: transformers.
Bumps the pip group with 1 update in the /allenai-Olmo-3-7B-Instruct/baseline directory: transformers.
Bumps the pip group with 1 update in the /allenai-Olmo-3-7B-Instruct/cpu directory: transformers.
Bumps the pip group with 1 update in the /allenai-Olmo-3-7B-Instruct/cuda directory: transformers.
Bumps the pip group with 1 update in the /allenai-Olmo-3-7B-Instruct/webgpu directory: transformers.
Bumps the pip group with 1 update in the /amd-AMD-OLMo-1B-SFT-DPO/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /codellama-CodeLlama-7b-Instruct-hf/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Llama-8B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Llama-8B/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/QNN directory: transformers.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/VitisAI directory: transformers.
Bumps the pip group with 2 updates in the /deepseek-ai-DeepSeek-R1-Distill-Qwen-7B/QNN directory: transformers and pytest.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Qwen-7B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /deepseek-ai-DeepSeek-R1-Distill-Qwen-7B/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /gpt-oss-20b/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-2-7b-chat-hf/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-2-7b-chat-hf/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-2-7b-hf/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-2-7b-hf/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.1-8B-Instruct/OpenVINO directory: transformers.
Bumps the pip group with 2 updates in the /meta-llama-Llama-3.1-8B-Instruct/QAIRT directory: transformers and tornado.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.1-8B-Instruct/QNN directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.1-8B-Instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.1-8B-Instruct/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.1-8B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.1-8B/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-1B-Instruct/QNN directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-1B-Instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-1B-Instruct/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-1B-Instruct/olive directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-1B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-3B-Instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-3B-Instruct/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Llama-3.2-3B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Meta-Llama-3-8B/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /meta-llama-Meta-Llama-3-8B/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3-mini-128k-instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3-mini-128k-instruct/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3-mini-4k-instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3-mini-4k-instruct/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3.5-mini-instruct/QNN directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3.5-mini-instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-3.5-mini-instruct/VitisAI directory: transformers.
Bumps the pip group with 5 updates in the /microsoft-Phi-4-mini-instruct/QAIRT directory:

Package	From	To
accelerate	1.13.0	1.14.0
cryptography	45.0.7	50.0.0
transformers	4.46.0	5.10.1
tornado	6.5.4	6.5.8
pytest	8.1.1	9.0.3

Bumps the pip group with 1 update in the /microsoft-Phi-4-mini-instruct/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-4-mini-instruct/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-4-mini-reasoning/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-4-mini-reasoning/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-4-reasoning/QAIRT directory: transformers.
Bumps the pip group with 1 update in the /microsoft-Phi-4-reasoning/QNN directory: transformers.
Bumps the pip group with 1 update in the /mistralai-Mistral-7B-Instruct-v0.1/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /mistralai-Mistral-7B-Instruct-v0.2/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /mistralai-Mistral-7B-Instruct-v0.2/VitisAI directory: transformers.
Bumps the pip group with 3 updates in the /mistralai-Mistral-7B-Instruct-v0.3/QAIRT directory: cryptography, transformers and tornado.
Bumps the pip group with 1 update in the /mistralai-Mistral-7B-Instruct-v0.3/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /mistralai-Mistral-7B-Instruct-v0.3/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /mistralai-Mistral-7B-v0.3/RyzenAI directory: transformers.
Bumps the pip group with 1 update in the /nvidia-Mistral-NeMo-12B-Instruct/baseline directory: transformers.
Bumps the pip group with 1 update in the /nvidia-Mistral-NeMo-12B-Instruct/cpu directory: transformers.
Bumps the pip group with 1 update in the /nvidia-Mistral-NeMo-12B-Instruct/cuda directory: transformers.
Bumps the pip group with 1 update in the /nvidia-Mistral-NeMo-12B-Instruct/webgpu directory: transformers.
Bumps the pip group with 1 update in the /openai-whisper-large-v3-turbo/cuda directory: transformers.
Bumps the pip group with 1 update in the /sd-legacy-stable-diffusion-v1-5/VitisAI directory: transformers.
Bumps the pip group with 1 update in the /tencent-HY-MT1.5-1.8B/baseline directory: transformers.
Bumps the pip group with 1 update in the /tencent-HY-MT1.5-1.8B/cpu directory: transformers.
Bumps the pip group with 1 update in the /tencent-HY-MT1.5-1.8B/cuda directory: transformers.
Bumps the pip group with 1 update in the /tencent-HY-MT1.5-1.8B/webgpu directory: transformers.

Updates accelerate from 1.6.0 to 1.14.0

Release notes

Sourced from accelerate's releases.

v1.14.0: AMD ROCm support, FSDP2 hardening

FSDP2 Improvements

This release brings a large batch of FSDP2 fixes and quality-of-life improvements: correct dtype handling on load, sharding of embeddings/norms, QLoRA crash prevention, and a more robust auto-wrap policy.

• Fsdp2 fully_shard embedding and norm by @​SunMarc in #4015
• Fix fsdp2 load full state dict dtype mismatch by @​SunMarc in #4021
• Fix region compilation fsdpv2 by @​SunMarc in #4022
• [FSDP2] Cast model to uniform dtype before fully_shard to fix mixed-dtype AssertionError by @​roycho96 in #3985
• [FSDP2] Auto-exclude non-floating frozen Params4bit from fully_shard to prevent QLoRA crash by @​roycho96 in #3987
• fix(FSDP2): auto-wrap policy ignoring _no_split_modules fallback by @​JohnGiorgi in #3999
• fix: use key-based matching in fsdp2_load_full_state_dict by @​roycho96 in #3982
• fix: add missing model_has_params4bit guard to fsdp2_load_full_state_dict call by @​roycho96 in #3981
• Fix to-fsdp2: drop REMOVED / NOT_YET_IMPLEMENTED FSDP1 keys instead of leaking them by @​lollinng in #4065
• Prevent double-wrapping models in prepare_model() by @​joshuaswanson in #3977

AMD ROCm support

Accelerate now works end-to-end on AMD ROCm devices. Thanks @​Abdennacer-Badaoui!

• Make accelerate work end-to-end on AMD ROCm by @​Abdennacer-Badaoui in #4025

Neuron

Further Neuron improvements to reduce recompilation and cover missing device cases.

• Add padded allgather and broadcast for Neuron devices to reduce recompilation by @​czkkkkkk in #4000
• fix: add missing neuron device case by @​michaelbenayoun in #4042

Quantization & Offloading

We improved offloading support for quantized models, including Torchao, int8, and tied-weight handling.

• Torchao offload by @​SunMarc in #3973
• Fix int8 offload hook detachment statistics restoration by @​jiqing-feng in #4044
• Fix keep_in_fp32_modules not working for tied weights in load_and_quantize_model by @​jiqing-feng in #4043
• Fix dtype_byte_size for FP8 fnuz / e8m0fnu dtypes by @​lollinng in #4063

Data Loading

• Feat: Support dynamic batch size in BatchSamplerShard with even_batches by @​yuxinyuan in #3969
• Fix iterable dataset sharding condition when n_shards == num_processes by @​SunMarc in #3958
• Fix implicit padding in split_between_processes when apply_padding=False and num_samples < num_processes by @​3manifold in #4052

Minor fixes

• [DeepSpeed] allow kernels flash-attn in SP by @​kashif in #3959
• Fix: Conditionally import torch.distributed.algorithms.join in accelerator.py by @​0xDELUXA in #3962
• Fix is_hf_initialized attribute by @​SunMarc in #3976
• feat(utils): add max reduction type by @​imstevenpmwork in #4027

... (truncated)

Commits

• beb0672 Release: v1.14.0
• a5cb37c Fix all tests (#4072)
• 14d7cdc Support dynamic batch size in BatchSamplerShard with even_batches (#3969)
• 77eb5cf fix(tracker): guard init_trackers and log against None kwargs (#4026)
• 379c8e0 fix(security): remediate workflow vulnerability in .github/workflows/build-do...
• 61e6420 Fix to-fsdp2: actually drop REMOVED / NOT_YET_IMPLEMENTED FSDP1 keys (#4065)
• b120c1f tests: Gracefully handle missing set_device for mps (#4028)
• 418015a test: add regression test for no_split_module_classes accepting set type (#4048)
• 3f1acf8 Bump the actions group with 8 updates (#4068)
• a2e5292 docs: add aggregate profiler memory example (#4054)
• Additional commits viewable in compare view

Updates cryptography from 49.0.0 to 50.0.0

Changelog

Sourced from cryptography's changelog.

50.0.0 - 2026-07-31

* **SECURITY ISSUE**:
  :func:`~cryptography.hazmat.primitives.serialization.pkcs7.pkcs7_decrypt_der`
  and its PEM and S/MIME variants no longer expose distinguishable errors or
  timing when unwrapping a ``RecipientInfo``'s ``encryptedKey``, which could
  act as a Bleichenbacher oracle for callers that decrypt untrusted messages.
  A random key is now substituted on failure, as described in :rfc:`3218`.
  Credit to **@X1AOxiang** for reporting the issue. **CVE-2026-69247**
* Deprecated Diffie-Hellman key exchange over finite fields (FFDH).
  Everything FFDH is deprecated, including the types in
  ``cryptography.hazmat.primitives.asymmetric.dh`` and loading FFDH keys or
  parameters with the key loading APIs. Users should migrate to a more
  modern key exchange algorithm.
* Added ``xof()`` class methods to
  :class:`~cryptography.hazmat.primitives.hashes.SHAKE128` and
  :class:`~cryptography.hazmat.primitives.hashes.SHAKE256` for constructing
  algorithm instances configured for use with
  :class:`~cryptography.hazmat.primitives.hashes.XOFHash`.
* The :mod:`X.509 verification <cryptography.x509.verification>` APIs are now
  considered stable and are subject to our API stability policy.
* Added the :doc:`/cobblestone` recipe, an implementation of the
  Cobblestone-128 and Cobblestone-256 instantiations of the `C2SP
  chunked-encryption specification
  <https://c2sp.org/chunked-encryption>`_ for streaming authenticated
  encryption of large messages.
* Parsing a Signed Certificate Timestamp list now rejects encodings that
  carry trailing bytes after the list or after an individual SCT, instead of
  silently ignoring them.
* Added support for using :class:`~cryptography.x509.Name` as a field type in
  the :doc:`/hazmat/asn1/index` module.
* Loading a public key or an EC private key now rejects DER where the
  ``subjectPublicKey`` (or EC ``publicKey``) ``BIT STRING`` declares a non-zero
  number of unused bits, instead of silently ignoring it.
* Parsing a CRL entry's ``InvalidityDate`` extension now rejects a
  ``GeneralizedTime`` that carries fractional seconds or another non-DER form,
  matching the strict encoding already required for every other X.509 time
  field.
* :func:`~cryptography.x509.ocsp.load_der_ocsp_request` and
  :func:`~cryptography.x509.ocsp.load_der_ocsp_response` now reject a request
  or response whose ``version`` field is not ``v1``, the only version defined
  by RFC 6960, matching the version validation already performed when loading
  certificates, CSRs and CRLs.
* :class:`~cryptography.hazmat.primitives.hashes.XOFHash` is now supported
  when building against AWS-LC.
* HMAC (and therefore PBKDF2-HMAC) with SHA-3 hashes is now supported when
  building against AWS-LC.
* Diffie-Hellman (:doc:`/hazmat/primitives/asymmetric/dh`) is now supported
  when building against AWS-LC.
</tr></table> 

... (truncated)

Commits

• dcb7050 Prepare for 50.0.0 release (#15372)
• 53fccd9 Don't leak how PKCS#7 encryptedKey decryption failed (#15369)
• d472f97 Add from __future__ import annotations to all src/ Python files (#15371)
• 908773d Bump downstream dependencies in CI (#15368)
• 2cc07cc Bump BoringSSL, OpenSSL, AWS-LC in CI (#15367)
• c94ede9 chore(deps): bump ruff from 0.16.0 to 0.16.1 (#15366)
• 67a8308 chore(deps): bump virtualenv from 21.7.0 to 21.7.1 (#15365)
• 95018ff Release the GIL in one-shot AEAD encrypt/decrypt (#15361)
• 6954733 Release the GIL during DH and DSA parameter generation (#15364)
• 6893b94 Import _serialization instead of serialization in x509/extensions (#15363)
• Additional commits viewable in compare view

Updates transformers from 4.51.3 to 5.10.1

Release notes

Sourced from transformers's releases.

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

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

Key differences from standard Gemma 4:

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

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

• who needs encoders? (#46385) by @​douglas-reid @​sgerrard @​vasqu @​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 | Paper

• Add Sapiens2 Model (#45919) by @​guarin in #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

• Add Deepseek-OCR-2 model (#45075) by @​thisisiron in #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

• feat: Add support for JetBrains' Mellum v2 code generation model (#46112) by @​shadeMe in #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) by @​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) by @​eustlb

... (truncated)

Commits

• 90c3ae5 Patch because we had to yank 5.10 because the release branch was not up to date
• 0bd94b3 v5.10.0
• 1423d22 who needs encoders? (#46385)
• 50eb20a Fix dsv4 dequant + tp/ep (#46378)
• 74464e8 Fix wrong changes produced by style/repo. check bot (#46371)
• 1b8ec34 Fix path traversal when saving Bark voice preset embeddings (#46237)
• e820678 Add Sapiens2 Model (#45919)
• 595721c Pass library_name/version to Hub calls via a shared HfApi (#46318)
• 0f0036c docs: update ACL Anthology URL in CITATION.cff (#46352)
• fa6c830 DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor (#45634)
• Additional commits viewable in compare view

Updates tornado from 6.5.5 to 6.5.8

Changelog

Sourced from tornado's changelog.

Release notes

.. toctree:: :maxdepth: 2

releases/v6.5.8 releases/v6.5.7 releases/v6.5.6 releases/v6.5.5 releases/v6.5.4 releases/v6.5.3 releases/v6.5.2 releases/v6.5.1 releases/v6.5.0 releases/v6.4.2 releases/v6.4.1 releases/v6.4.0 releases/v6.3.3 releases/v6.3.2 releases/v6.3.1 releases/v6.3.0 releases/v6.2.0 releases/v6.1.0 releases/v6.0.4 releases/v6.0.3 releases/v6.0.2 releases/v6.0.1 releases/v6.0.0 releases/v5.1.1 releases/v5.1.0 releases/v5.0.2 releases/v5.0.1 releases/v5.0.0 releases/v4.5.3 releases/v4.5.2 releases/v4.5.1 releases/v4.5.0 releases/v4.4.3 releases/v4.4.2 releases/v4.4.1 releases/v4.4.0 releases/v4.3.0 releases/v4.2.1 releases/v4.2.0 releases/v4.1.0 releases/v4.0.2 releases/v4.0.1 releases/v4.0.0 releases/v3.2.2

... (truncated)

Commits

• a55abe3 Merge pull request #3704 from bdarnell/security-6.5.8
• fc79488 docs: add additional credit to release notes
• 7b01763 Fix test_strip_headers_on_redirect's URL-embedded-credentials cases
• d72fff8 release notes and version bump for 6.5.8
• b168818 auth: Formally deprecated OpenIDMixin
• da28476 web: Also check for semicolons in deprecated mixed-case cookie args
• 8d6363e httputil: Enforce a new limit on the number of arguments in a request
• de85b3f httputil: Apply multipart max_parts limit earlier
• 48fc2d4 Merge pull request #3633 from bdarnell/curl-reset-65
• 4ae1ddd Release notes and version bump for 6.5.7
• Additional commits viewable in compare view

Updates transformers from 4.52.4 to 5.10.1

Release notes

Sourced from transformers's releases.

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

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

Key differences from standard Gemma 4:

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

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

• who needs encoders? (#46385) by @​douglas-reid @​sgerrard @​vasqu @​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 | Paper

• Add Sapiens2 Model (#45919) by @​guarin in #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

• Add Deepseek-OCR-2 model (#45075) by @​thisisiron in #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

• feat: Add support for JetBrains' Mellum v2 code generation model (#46112) by @​shadeMe in #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) by @​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) by @​eustlb

... (truncated)

Commits

• 90c3ae5 Patch because we had to yank 5.10 because the release branch was not up to date
• 0bd94b3 v5.10.0
• 1423d22 who needs encoders? (#46385)
• 50eb20a Fix dsv4 dequant + tp/ep (#46378)
• 74464e8 Fix wrong changes produced by style/repo. check bot (#46371)
• 1b8ec34 Fix path traversal when saving Bark voice preset embeddings (#46237)
• e820678 Add Sapiens2 Model (#45919)
• 595721c Pass library_name/version to Hub calls via a shared HfApi (#46318)
• 0f0036c docs: update ACL Anthology URL in CITATION.cff (#46352)
• fa6c830 DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor (#45634)
• Additional commits viewable in compare view

Updates transformers from 4.52.4 to 5.10.1

Release notes

Sourced from transformers's releases.

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

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

Key differences from standard Gemma 4:

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

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

• who needs encoders? (#46385) by @​douglas-reid @​sgerrard @​vasqu @​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 | Paper

• Add Sapiens2 Model (#45919) by @​guarin in #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

• Add Deepseek-OCR-2 model (#45075) by @​thisisiron in #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

• feat: Add support for JetBrains' Mellum v2 code generation model (#46112) by @​shadeMe in #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) by @​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) by @​eustlb

... (truncated)

Commits

• 90c3ae5 Patch because we had to yank 5.10 because the release branch was not up to date
• 0bd94b3 v5.10.0
• 1423d22 who needs encoders? (#46385)
• 50eb20a Fix dsv4 dequant + tp/ep (#46378)
• 74464e8 Fix wrong changes produced by style/repo. check bot (#46371)
• 1b8ec34 Fix path traversal when saving Bark voice preset embeddings (#46237)
• e820678 Add Sapiens2 Model (#45919)
• 595721c Pass library_name/version to Hub calls via a shared HfApi (#46318)
• 0f0036c docs: update ACL Anthology URL in CITATION.cff (#46352)
• fa6c830 DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor (#45634)
• Additional commits viewable in compare view

Updates transformers from 4.52.4 to 5.10.1

Release notes

Sourced from transformers's releases.

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

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

Key differences from standard Gemma 4:

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

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

• who needs encoders? (#46385) by @​douglas-reid @​sgerrard @​vasqu @​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 | Paper

• Add Sapiens2 Model (#45919) by @​guarin in #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

• Add Deepseek-OCR-2 model (#45075) by @​thisisiron in #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

• feat: Add support for JetBrains' Mellum v2 code generation model (#46112) by @​shadeMe in #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) by @​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) by @​eustlb

... (truncated)

Commits

• 90c3ae5 Patch because we had to yank 5.10 because the release branch was not up to date
• 0bd94b3 v5.10.0
• 1423d22 who needs encoders? (#46385)
• 50eb20a Fix dsv4 dequant + tp/ep (#46378)
• 74464e8 Fix wrong changes produced by style/repo. check bot (#46371)
• 1b8ec34 Fix path traversal when saving Bark voice preset embeddings (#46237)
• e820678 Add Sapiens2 Model (#45919)
• 595721c Pass library_name/version to Hub calls via a shared HfApi (#46318)
• 0f0036c docs: update ACL Anthology URL in CITATION.cff (#46352)
• fa6c830 Deep...

  Description has been truncated

[Copilot]

🔵 Needs a closer look

It performs large-scale dependency upgrades (including a major Transformers 4→5 jump) across many environments and includes at least one suspicious version downgrade/naming inconsistency that needs human validation.

Pull request overview

Updates pinned Python dependencies across many model/target-specific recipe environments, primarily standardizing on newer transformers (and related bumps in a few central .aitk requirement sets).

Changes:

• Bump transformers pins across numerous recipe requirement files to 5.10.1.
• Update select .aitk requirement sets to newer accelerate, cryptography, tornado, and pytest versions.
• Refresh some platform/accelerator-specific requirement snapshots (QAIRT/QNN/RyzenAI/VitisAI/OpenVINO/WebGPU/CUDA/CPU).

File summaries

File	Description
tencent-HY-MT1.5-1.8B/webgpu/requirements.txt	Bump transformers pin to 5.10.1
tencent-HY-MT1.5-1.8B/cuda/requirements.txt	Bump transformers pin to 5.10.1
tencent-HY-MT1.5-1.8B/cpu/requirements.txt	Bump transformers pin to 5.10.1
tencent-HY-MT1.5-1.8B/baseline/requirements.txt	Bump transformers pin to 5.10.1
sd-legacy-stable-diffusion-v1-5/VitisAI/requirements_vitisai_sd.txt	Bump transformers pin to 5.10.1
Qwen-Qwen3.5-2B/webgpu/requirements.txt	Bump transformers pin to 5.10.1
Qwen-Qwen3.5-2B/cuda/requirements.txt	Bump transformers pin to 5.10.1
Qwen-Qwen3.5-2B/cpu/requirements.txt	Bump transformers pin to 5.10.1
openai-whisper-large-v3-turbo/cuda/eval-requirements.txt	Bump transformers pin to 5.10.1
nvidia-Mistral-NeMo-12B-Instruct/webgpu/requirements.txt	Bump transformers pin to 5.10.1
nvidia-Mistral-NeMo-12B-Instruct/cuda/requirements.txt	Bump transformers pin to 5.10.1
nvidia-Mistral-NeMo-12B-Instruct/cpu/requirements.txt	Bump transformers pin to 5.10.1
nvidia-Mistral-NeMo-12B-Instruct/baseline/requirements.txt	Bump transformers pin to 5.10.1
mistralai-Mistral-7B-v0.3/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
mistralai-Mistral-7B-Instruct-v0.3/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
mistralai-Mistral-7B-Instruct-v0.3/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
mistralai-Mistral-7B-Instruct-v0.3/QAIRT/requirements.txt	Bump cryptography, tornado, transformers pins
mistralai-Mistral-7B-Instruct-v0.2/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
mistralai-Mistral-7B-Instruct-v0.2/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
mistralai-Mistral-7B-Instruct-v0.1/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-reasoning/QNN/requirements.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-reasoning/QAIRT/requirements.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-mini-reasoning/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-mini-reasoning/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-mini-instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-mini-instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-4-mini-instruct/QAIRT/requirements.txt	Bump accelerate, cryptography, pytest, tornado, transformers pins
microsoft-Phi-3.5-mini-instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-3.5-mini-instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-3.5-mini-instruct/QNN/requirements.txt	Bump transformers pin to 5.10.1
microsoft-Phi-3-mini-4k-instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-3-mini-4k-instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-3-mini-128k-instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
microsoft-Phi-3-mini-128k-instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Meta-Llama-3-8B/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Meta-Llama-3-8B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-3B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-3B-Instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-3B-Instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-1B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-1B-Instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-1B-Instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-1B-Instruct/QNN/requirements.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.2-1B-Instruct/olive/requirements.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.1-8B/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.1-8B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.1-8B-Instruct/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.1-8B-Instruct/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.1-8B-Instruct/QNN/requirements.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-3.1-8B-Instruct/QAIRT/requirements.txt	Bump tornado and transformers pins
meta-llama-Llama-3.1-8B-Instruct/OpenVINO/requirements.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-2-7b-hf/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-2-7b-hf/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-2-7b-chat-hf/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
meta-llama-Llama-2-7b-chat-hf/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
gpt-oss-20b/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Qwen-7B/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Qwen-7B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Qwen-7B/QNN/requirements.txt	Bump pytest and transformers pins
deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Qwen-1.5B/QNN/requirements.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Llama-8B/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
deepseek-ai-DeepSeek-R1-Distill-Llama-8B/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
codellama-CodeLlama-7b-Instruct-hf/RyzenAI/requirements_ryzenai_llm.txt	Bump transformers pin to 5.10.1
amd-AMD-OLMo-1B-SFT-DPO/VitisAI/requirements_vitisai_llm.txt	Bump transformers pin to 5.10.1
allenai-Olmo-3-7B-Instruct/webgpu/requirements.txt	Bump transformers pin to 5.10.1
allenai-Olmo-3-7B-Instruct/cuda/requirements.txt	Bump transformers pin to 5.10.1
allenai-Olmo-3-7B-Instruct/cpu/requirements.txt	Bump transformers pin to 5.10.1
allenai-Olmo-3-7B-Instruct/baseline/requirements.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-WinMLCLI.txt	Bump cryptography and transformers pins
.aitk/requirements/requirements-WCR.txt	Bump tornado and transformers pins
.aitk/requirements/requirements-WCR-SD.txt	Bump accelerate pin
.aitk/requirements/requirements-WCR-SAM.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-QNN.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-QNN_LLM.txt	Bump tornado pin
.aitk/requirements/requirements-Profiling.txt	Bump tornado pin
.aitk/requirements/requirements-NvidiaGPU.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-NvidiaGPU-Qwen3.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-NvidiaGPU-GptqModel.txt	Bump accelerate and transformers pins
.aitk/requirements/requirements-NvidiaGPU-AutoGptq.txt	Bump accelerate pin
.aitk/requirements/requirements-NvidiaGPU-AutoAwq.txt	Bump accelerate pin
.aitk/requirements/requirements-IntelNPU.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-IntelNPU-WP.txt	Bump transformers pin to 5.10.1
.aitk/requirements/requirements-IntelNPU-SD.txt	Bump accelerate pin
.aitk/requirements/Intel/Test_py3.12.9.txt	Bump tornado and transformers pins
.aitk/requirements/Intel/Test_py3.12.9-Transformers4.49.txt	Bump transformers pin to 5.10.1
.aitk/requirements/General/CUDA_py3.12.9.txt	Bump transformers pin to 5.10.1
.aitk/requirements/General/CPU_py3.12.9.txt	Bump transformers pin to 5.10.1
.aitk/requirements/General/CPU_py3.12.9-SD.txt	Bump accelerate pin
.aitk/requirements/General/CPU_py3.12.9-SAM.txt	Bump transformers pin to 5.10.1
.aitk/requirements/General/CPU_py3.12.9-genai.txt	Set transformers pin to 5.10.1 (note: differs from prior 5.14.1)
.aitk/requirements/AMD/Quark_py3.12.13.txt	Bump accelerate and transformers pins

Review details

• Files reviewed: 93/93 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[dependabot]

Dependabot tried to update this pull request, but something went wrong. We're looking into it, but in the meantime you can retry the update by commenting @dependabot recreate.

[dependabot]

Dependabot tried to update this pull request, but something went wrong. We're looking into it, but in the meantime you can retry the update by commenting @dependabot recreate.

[dependabot]

Dependabot tried to update this pull request, but something went wrong. We're looking into it, but in the meantime you can retry the update by commenting @dependabot recreate.