Title: Bump the pip group across 8 directories with 7 updates by dependabot[bot] · Pull Request #2 · haykeminyan/pythoncode-tutorials · GitHub
Open Graph Title: Bump the pip group across 8 directories with 7 updates by dependabot[bot] · Pull Request #2 · haykeminyan/pythoncode-tutorials
X Title: Bump the pip group across 8 directories with 7 updates by dependabot[bot] · Pull Request #2 · haykeminyan/pythoncode-tutorials
Description: Bumps the pip group with 2 updates in the /general/detect-fraudulent-transactions directory: kafka-python and flask. Bumps the pip group with 1 update in the /handling-pdf-files/pdf-ocr directory: pillow. Bumps the pip group with 1 update in the /machine-learning/asr directory: transformers. Bumps the pip group with 1 update in the /machine-learning/nlp/machine-translation directory: transformers. Bumps the pip group with 1 update in the /machine-learning/nlp/pretraining-bert directory: transformers. Bumps the pip group with 1 update in the /machine-learning/nlp/speech-recognition-transformers directory: transformers. Bumps the pip group with 2 updates in the /web-programming/accounting-app directory: django and sqlparse. Bumps the pip group with 2 updates in the /web-programming/bookshop-crud-app-flask-fullstack directory: flask and werkzeug. Updates kafka-python from 2.0.2 to 2.3.2 Release notes Sourced from kafka-python's releases. 2.3.2 (Jun 3, 2026) Fixes kafka.conn: Improve error handling for sasl authenticate mechanisms kafka.net: Validate SASL/SCRAM iterations (#3026) KafkaProtocol: validate network frame size (backport of #3019) Clear _callbacks/_errbacks list when future is_done to avoid reference cycles (#2891) 2.3.1 (Apr 9, 2026) Fixes Fix TaggedFields value encoding; add test coverage (#2725) Fix zstd multi-frame decompression failure (#2717) Fix KeyError in KafkaConsumer.committed() (#2710) Fix VarInt/VarLong encoding; move tests to test/protocol/ (#2706) Fix Fetcher._fetch_offsets_by_times retry handling (#2833) Fixes to support integration testing with external KAFKA_URI (#2838) Minor py2 fixes: consumer integration test; dont pip install python-snappy 2.3.0 (Nov 20, 2025) 2.3.x will be the last release branch with python2 support! CLI python -m cli interfaces for kafka.admin, kafka.consumer, kafka.producer (#2650) Producer KIP-654: Abort transaction with pending data with TransactionAbortedError (#2662) KafkaProducer: Handle UnknownProducerIdError (#2663) KIP-467: Augment ProduceResponse error messaging for specific culprit records (#2661) Consumer KIP-345: Consumer group static membership (#2625) KIP-207: Add ListOffsetsRequest v5 / handle OffsetNotAvailableError (#2657) Fetcher: Add missing argument in debug log (#2665) AdminClient KIP-430: Return Authorized Operations in Describe Responses (#2656) Add send_request() and send_requests() to KafkaAdminClient (#2649) Maintenance Remove old/unused kafka.protocol.pickle (#2653) Switch protocol code to getattr/setattr from dict (#2654) ... (truncated) Changelog Sourced from kafka-python's changelog. 2.3.2 (Jun 3, 2026) ################### Fixes kafka.conn: Improve error handling for sasl authenticate mechanisms kafka.net: Validate SASL/SCRAM iterations (#3026) KafkaProtocol: validate network frame size (backport of #3019) Clear _callbacks/_errbacks list when future is_done to avoid reference cycles (#2891) 2.3.1 (Apr 9, 2026) ################### Fixes Fix TaggedFields value encoding; add test coverage (#2725) Fix zstd multi-frame decompression failure (#2717) Fix KeyError in KafkaConsumer.committed() (#2710) Fix VarInt/VarLong encoding; move tests to test/protocol/ (#2706) Fix Fetcher._fetch_offsets_by_times retry handling (#2833) Fixes to support integration testing with external KAFKA_URI (#2838) Minor py2 fixes: consumer integration test; dont pip install python-snappy 2.3.0 (Nov 20, 2025) #################### CLI python -m cli interfaces for kafka.admin, kafka.consumer, kafka.producer (#2650) Producer KIP-654: Abort transaction with pending data with TransactionAbortedError (#2662) KafkaProducer: Handle UnknownProducerIdError (#2663) KIP-467: Augment ProduceResponse error messaging for specific culprit records (#2661) Add transactional_id to KafkaProducer Keyword Arguments docstring Consumer KIP-345: Consumer group static membership (#2625) KIP-207: Add ListOffsetsRequest v5 / handle OffsetNotAvailableError (#2657) Fetcher: Add missing argument in debug log (#2665) AdminClient KIP-430: Return Authorized Operations in Describe Responses (#2656) Add send_request() and send_requests() to KafkaAdminClient (#2649) ... (truncated) Commits 054b570 Patch Release 2.3.2 0efe08b kafka.conn: Improve error handling for sasl authenticate mechanisms 74400d7 kafka.net: Validate SASL/SCRAM iterations (#3026) 9f92d0f KafkaProtocol: validate network frame size (backport of #3019) 79410e7 Clear _callbacks/_errbacks list when future is_done to avoid reference cycles... 8e50f65 Enable manual workflow runs for Python Package 62feaab Patch Release 2.3.1 8037f81 Fixes to support integration testing with external KAFKA_URI (#2838) 3479de9 python-snappy py3 only 2c72167 Fix TaggedFields value encoding; add test coverage (#2725) Additional commits viewable in compare view Updates flask from 1.1.2 to 3.1.3 Release notes Sourced from flask's releases. 3.1.3 This is the Flask 3.1.3 security fix release, which fixes a security issue but does not otherwise change behavior and should not result in breaking changes compared to the latest feature release. PyPI: https://pypi.org/project/Flask/3.1.3/ Changes: https://flask.palletsprojects.com/page/changes/#version-3-1-3 The session is marked as accessed for operations that only access the keys but not the values, such as in and len. GHSA-68rp-wp8r-4726 3.1.2 This is the Flask 3.1.2 fix release, which fixes bugs but does not otherwise change behavior and should not result in breaking changes compared to the latest feature release. PyPI: https://pypi.org/project/Flask/3.1.2/ Changes: https://flask.palletsprojects.com/page/changes/#version-3-1-2 Milestone: https://github.com/pallets/flask/milestone/38?closed=1 stream_with_context does not fail inside async views. #5774 When using follow_redirects in the test client, the final state of session is correct. #5786 Relax type hint for passing bytes IO to send_file. #5776 3.1.1 This is the Flask 3.1.1 fix release, which fixes bugs but does not otherwise change behavior and should not result in breaking changes compared to the latest feature release. PyPI: https://pypi.org/project/Flask/3.1.1/ Changes: https://flask.palletsprojects.com/en/stable/changes/#version-3-1-1 Milestone https://github.com/pallets/flask/milestone/36?closed=1 Fix signing key selection order when key rotation is enabled via SECRET_KEY_FALLBACKS. GHSA-4grg-w6v8-c28g Fix type hint for cli_runner.invoke. #5645 flask --help loads the app and plugins first to make sure all commands are shown. #5673 Mark sans-io base class as being able to handle views that return AsyncIterable. This is not accurate for Flask, but makes typing easier for Quart. #5659 3.1.0 This is the Flask 3.1.0 feature release. A feature release may include new features, remove previously deprecated code, add new deprecations, or introduce potentially breaking changes. We encourage everyone to upgrade, and to use a tool such as pip-tools to pin all dependencies and control upgrades. Test with warnings treated as errors to be able to adapt to deprecation warnings early. PyPI: https://pypi.org/project/Flask/3.1.0/ Changes: https://flask.palletsprojects.com/en/stable/changes/#version-3-1-0 Milestone: https://github.com/pallets/flask/milestone/33?closed=1 Drop support for Python 3.8. #5623 Update minimum dependency versions to latest feature releases. Werkzeug >= 3.1, ItsDangerous >= 2.2, Blinker >= 1.9. #5624, #5633 Provide a configuration option to control automatic option responses. #5496 Flask.open_resource/open_instance_resource and Blueprint.open_resource take an encoding parameter to use when opening in text mode. It defaults to utf-8. #5504 Request.max_content_length can be customized per-request instead of only through the MAX_CONTENT_LENGTH config. Added MAX_FORM_MEMORY_SIZE and MAX_FORM_PARTS config. Added documentation about resource limits to the security page. #5625 Add support for the Partitioned cookie attribute (CHIPS), with the SESSION_COOKIE_PARTITIONED config. #5472 -e path takes precedence over default .env and .flaskenv files. load_dotenv loads default files in addition to a path unless load_defaults=False is passed. #5628 Support key rotation with the SECRET_KEY_FALLBACKS config, a list of old secret keys that can still be used for unsigning. Extensions will need to add support. #5621 Fix how setting host_matching=True or subdomain_matching=False interacts with SERVER_NAME. Setting SERVER_NAME no longer restricts requests to only that domain. #5553 Request.trusted_hosts is checked during routing, and can be set through the TRUSTED_HOSTS config. #5636 3.0.3 ... (truncated) Changelog Sourced from flask's changelog. Version 3.1.3 Released 2026-02-18 The session is marked as accessed for operations that only access the keys but not the values, such as in and len. :ghsa:68rp-wp8r-4726 Version 3.1.2 Released 2025-08-19 stream_with_context does not fail inside async views. :issue:5774 When using follow_redirects in the test client, the final state of session is correct. :issue:5786 Relax type hint for passing bytes IO to send_file. :issue:5776 Version 3.1.1 Released 2025-05-13 Fix signing key selection order when key rotation is enabled via SECRET_KEY_FALLBACKS. :ghsa:4grg-w6v8-c28g Fix type hint for cli_runner.invoke. :issue:5645 flask --help loads the app and plugins first to make sure all commands are shown. :issue:5673 Mark sans-io base class as being able to handle views that return AsyncIterable. This is not accurate for Flask, but makes typing easier for Quart. :pr:5659 Version 3.1.0 Released 2024-11-13 Drop support for Python 3.8. :pr:5623 Update minimum dependency versions to latest feature releases. Werkzeug >= 3.1, ItsDangerous >= 2.2, Blinker >= 1.9. :pr:5624,5633 Provide a configuration option to control automatic option responses. :pr:5496 Flask.open_resource/open_instance_resource and Blueprint.open_resource take an encoding parameter to use when opening in text mode. It defaults to utf-8. :issue:5504 Request.max_content_length can be customized per-request instead of only through the MAX_CONTENT_LENGTH config. Added ... (truncated) Commits 22d9247 release version 3.1.3 089cb86 Merge commit from fork c17f379 request context tracks session access 27be933 start version 3.1.3 4e652d3 Abort if the instance folder cannot be created (#5903) 3d03098 Abort if the instance folder cannot be created 407eb76 document using gevent for async (#5900) ac5664d document using gevent for async 4f79d5b Increase required flit_core version to 3.11 (#5865) fe3b215 Increase required flit_core version to 3.11 Additional commits viewable in compare view Updates pillow from 8.0.1 to 13.0.0 Release notes Sourced from pillow's releases. 13.0.0 https://pillow.readthedocs.io/en/stable/releasenotes/13.0.0.html Removals Remove undocumented GifImagePlugin._save_netpbm #9901 [@akx] Remove ExifTags.IFD.Makernote #9850 [@radarhere] Drop support for Python 3.10 #9750 [@hugovk] Remove deprecations for Pillow 13.0.0 #9751 [@radarhere] Deprecations Deprecate IM image format #9849 [@radarhere] Deprecate ImageQt align8to32() in favour of raw encoder stride #9940 [@akx] Deprecate JpegImageFile.load_djpeg #9924 [@akx] Documentation Update release notes for Pillow 13.0.0 #10223 [@hugovk] Correct the ImageOps.fit centering corner documentation #10118 [@FanWu-ai] Update Tidelift links #10102 [@radarhere] Correct bullet point syntax in release notes #10069 [@radarhere] Update ImageDraw signatures in docs #10065 [@ZainnQureshii] Update macOS tested Pillow versions #10048 [@hugovk] Add pillow-wmf and pillow-zx-spectrum to third-party plugins list #10037 [@bitplane] Delete CLAUDE.md #10027 [@hugovk] Add pillow-lunasvg to third-party plugins list #10016 [@Maksim-Burtsev] Extend security policy #10018 [@hugovk] Update Color3DLUT description #10004 [@radarhere] Document the internal image distinctness constraints #9752 [@akx] Update macOS tested Pillow versions #10000 [@radarhere] Link security policy in project metadata #9980 [@hugovk] Image.new(): correct docstring and comment about uninitialized memory #9960 [@akx] Add AGENTS.md #9582 [@aclark4life] Add release notes for #8811 #9937 [@radarhere] Fixed typo #9923 [@radarhere] Fixed typo #9922 [@radarhere] C interface is internal #9879 [@radarhere] Correct function references #9895 [@radarhere] Revert move from python-pillow.org to python-pillow.github.io #9830 [@aclark4life] Add additional CVE numbers to 12.3.0 release notes #9777 [@radarhere] Remove remaining :rtype:s from docstrings #9742 [@akx] Update macOS tested Pillow versions #9759 [@radarhere] Dependencies Update dependency cibuildwheel to v4.3.1 #10175 [@renovate[bot]] Update dependency cibuildwheel to v4.3.0 #10132 [@renovate[bot]] Update dependency harfbuzz to v14.5.0 #10108 [@renovate[bot]] Update github-actions #10111 [@renovate[bot]] ... (truncated) Changelog Sourced from pillow's changelog. Changelog (Pillow) 11.1.0 and newer See GitHub Releases: https://github.com/python-pillow/Pillow/releases 11.0.0 (2024-10-15) Update licence to MIT-CMU #8460 [hugovk] Conditionally define ImageCms type hint to avoid requiring core #8197 [radarhere] Support writing LONG8 offsets in AppendingTiffWriter #8417 [radarhere] Use ImageFile.MAXBLOCK when saving TIFF images #8461 [radarhere] Do not close provided file handles with libtiff when saving #8458 [radarhere] Support ImageFilter.BuiltinFilter for I;16* images #8438 [radarhere] Use ImagingCore.ptr instead of ImagingCore.id #8341 [homm, radarhere, hugovk] Updated EPS mode when opening images without transparency #8281 [Yay295, radarhere] Use transparency when combining P frames from APNGs #8443 [radarhere] Support all resampling filters when resizing I;16* images #8422 [radarhere] Free memory on early return #8413 [radarhere] Cast int before potentially exceeding INT_MAX #8402 [radarhere] ... (truncated) Commits 293218a 13.0.0 version bump 262e8e4 Update release notes for Pillow 13.0.0 (#10223) af56ace Validate TIFF tag values against libtiff's own field definition before TIFFSe... 570df03 Validate TIFF tag values against libtiff's own field definition before TIFFSe... 53318b5 Take libtiff scanline geometry tags from the image, not tiffinfo/exif (#10226) 0a87417 Bound retained TIFF tag data by stream size (#10225) 11582e5 Fix heap buffer overflow in _tiffWriteProc when saving TIFF to BytesIO (#10224) 8501adc Fix test ec5f1cb Take libtiff scanline geometry tags from the image, not tiffinfo/exif 4b63dff Share TIFF tag data budget across frames Additional commits viewable in compare view Updates transformers from 4.28.1 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 4.12.4 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 4.18.0 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 4.28.1 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 Description has been truncated
Open Graph Description: Bumps the pip group with 2 updates in the /general/detect-fraudulent-transactions directory: kafka-python and flask. Bumps the pip group with 1 update in the /handling-pdf-files/pdf-ocr directory: ...
X Description: Bumps the pip group with 2 updates in the /general/detect-fraudulent-transactions directory: kafka-python and flask. Bumps the pip group with 1 update in the /handling-pdf-files/pdf-ocr directory: ...
Opengraph URL: https://github.com/haykeminyan/pythoncode-tutorials/pull/2
X: @github
Domain: github.com
| react-profiling | 0 |
| react-import-map | react,react-dom,react-dom/client,react-dom/profiling,react-is,react/compiler-runtime,react/jsx-dev-runtime,react/jsx-runtime,scheduler@b8488ee2c322 |
| route-pattern | /:user_id/:repository/pull/:id/files(.:format) |
| route-controller | pull_requests |
| route-action | files |
| fetch-nonce | v2:0cfa2eb8-7bd0-762a-1e06-ce5e1b18c760 |
| current-catalog-service-hash | ae870bc5e265a340912cde392f23dad3671a0a881730ffdadd82f2f57d81641b |
| request-id | EB8C:21BBCD:2EFF89:3CC127:6AFF8973 |
| html-safe-nonce | 1c1f363d8acf7a4291c7e3abce5f5e4cf17cc4916900fdf5f3175d878a4527f7 |
| visitor-payload | eyJyZWZlcnJlciI6IiIsInJlcXVlc3RfaWQiOiJFQjhDOjIxQkJDRDoyRUZGODk6M0NDMTI3OjZBRkY4OTczIiwidmlzaXRvcl9pZCI6IjE1OTYxNDMyNjM1NTc4NDc0MTEiLCJyZWdpb25fZWRnZSI6ImlhZCIsInJlZ2lvbl9yZW5kZXIiOiJpYWQifQ== |
| visitor-hmac | c637302b8ab48ff98156087224b776501480cbd4ceb64d830f6708ca01df1685 |
| hovercard-subject-tag | pull_request:4995520544 |
| github-keyboard-shortcuts | repository,pull-request-list,pull-request-conversation,pull-request-files-changed,copilot |
| google-site-verification | Apib7-x98H0j5cPqHWwSMm6dNU4GmODRoqxLiDzdx9I |
| octolytics-url | https://collector.github.com/github/collect |
| analytics-location | / |
| fb:app_id | 1401488693436528 |
| apple-itunes-app | app-id=1477376905, app-argument=https://github.com/haykeminyan/pythoncode-tutorials/pull/2/files |
| twitter:image | https://avatars.githubusercontent.com/in/29110?s=400&v=4 |
| twitter:card | summary_large_image |
| og:image | https://avatars.githubusercontent.com/in/29110?s=400&v=4 |
| og:image:alt | Bumps the pip group with 2 updates in the /general/detect-fraudulent-transactions directory: kafka-python and flask. Bumps the pip group with 1 update in the /handling-pdf-files/pdf-ocr directory: ... |
| og:site_name | GitHub |
| og:type | object |
| hostname | github.com |
| expected-hostname | github.com |
| None | 587ad99a3e443b2e1cb0d3df059ec49ab54389861f13310766dfdeb615fed90d |
| turbo-cache-control | no-preview |
| diff-view | unified |
| go-import | github.com/haykeminyan/pythoncode-tutorials git https://github.com/haykeminyan/pythoncode-tutorials.git |
| octolytics-dimension-user_id | 43892880 |
| octolytics-dimension-user_login | haykeminyan |
| octolytics-dimension-repository_id | 950304910 |
| octolytics-dimension-repository_nwo | haykeminyan/pythoncode-tutorials |
| octolytics-dimension-repository_public | true |
| octolytics-dimension-repository_is_fork | true |
| octolytics-dimension-repository_parent_id | 199449624 |
| octolytics-dimension-repository_parent_nwo | x4nth055/pythoncode-tutorials |
| octolytics-dimension-repository_network_root_id | 199449624 |
| octolytics-dimension-repository_network_root_nwo | x4nth055/pythoncode-tutorials |
| turbo-body-classes | logged-out env-production page-responsive full-width |
| disable-turbo | true |
| browser-stats-url | https://api.github.com/_private/browser/stats |
| browser-errors-url | https://api.github.com/_private/browser/errors |
| release | 5bb0fb68d64881d1344f5863d809515d496801e5 |
| ui-target | full |
| theme-color | #1e2327 |
| color-scheme | light dark |
Links:
Viewport: width=device-width