Bump the pip group across 8 directories with 3 updates

[dependabot]

⚠️ Dependabot is rebasing this PR ⚠️

Rebasing might not happen immediately, so don't worry if this takes some time.

Note: if you make any changes to this PR yourself, they will take precedence over the rebase.

---

Bumps the pip group with 1 update in the /examples/pytorch/diffusion_model/diffusers/wan directory: gdown.
Bumps the pip group with 1 update in the /examples/pytorch/nlp/huggingface_models/language-modeling/quantization/auto_round/deepseek directory: transformers.
Bumps the pip group with 1 update in the /examples/pytorch/nlp/huggingface_models/language-modeling/quantization/auto_round/llama3 directory: transformers.
Bumps the pip group with 1 update in the /examples/pytorch/nlp/huggingface_models/language-modeling/quantization/llm_qat directory: transformers.
Bumps the pip group with 1 update in the /examples/pytorch/nlp/huggingface_models/language-modeling/quantization/smooth_quant directory: torch.
Bumps the pip group with 1 update in the /examples/pytorch/nlp/huggingface_models/language-modeling/quantization/static_quant/pt2e directory: transformers.
Bumps the pip group with 1 update in the /examples/pytorch/nlp/huggingface_models/language-modeling/quantization/transformers/weight_only/text-generation directory: torch.
Bumps the pip group with 1 update in the /examples/tensorflow/nlp/large_language_models/quantization/ptq/smoothquant directory: transformers.

Updates gdown from 4.7.3 to 5.2.2

Release notes

Sourced from gdown's releases.

v5.2.2

Security

• Fix path traversal vulnerability in extractall() that allowed zip/tar archives with ../ entries to write files outside the target directory (GHSA-76hw-p97h-883f)
• Reject symlinks, hardlinks, and special files in tar archives
• Use Python 3.12+ filter="data" for safe tar extraction when available
• Sanitize filenames from HTTP responses and URLs to prevent path traversal via /, , .., and null bytes
• Sanitize root folder name in download_folder() before building directory paths

v5.2.1

Fixes

• cached_download: Verify file hash before moving to final location instead of after (wkentaro/gdown#417)
  • Previously, the hash was verified after moving to the final path, which could leave corrupted files in place if hash verification failed
• download: Fix speed limit throttling logic to use independent byte counter instead of pbar.n (wkentaro/gdown#407)
  • The speed limiter now correctly tracks downloaded bytes independently from the progress bar
  • Also fixes unnecessary sleep when resuming downloads with speed limit (since pbar.n includes start_size from resumed downloads)

Chores

• Fix missing space in --output help text (wkentaro/gdown#398)
• Fix concatenated string literal in extractall.py error message

v5.2.0

🚀 Features

• Feature: preserve GDrive's last modified time when downloading #351 (Thanks @​moulins)
• Feature: allow folders to be continued more usefully #288 (Thanks @​achadwick)

✨ Enhancement

• Show resume progress bar from the middle #361

🐛 Fixes

• Use importlib.metadata to get the version number from the installed package #319
• BUG: Download folder with "skip_download=True" creates an output folder #321 (Thanks @​o-laurent)

💬 Other

• Fix typos in README.md #323 (Thanks @​amorehead)
• Run GitHub Actions on macOS, Ubuntu, and Windows #324
• Support Windows and macOS in tests #359
• Typo in arg help text #343 (Thanks @​frosit)
• Trivial: Update the documentation for the output param #347 (Thanks @​abouelkhair5)

---

• To connect, check my Twitter.
• To support, check Sponsoring Page.

... (truncated)

Commits

• af569fc fix: prevent path traversal in archive extraction and filename handling
• 7f4cb68 Merge pull request #417 from wkentaro/fix_cached_download_verify_file_hash_be...
• 9697a53 fix(cached_download): verify file hash before moving to final location
• c7c1b9d Merge pull request #407 from wkentaro/fix_speed_limit
• 940bd40 Fix throttling logic to use independent byte counter instead of pbar.n
• a55ce67 Fix unnecessary sleep when resuming downloads with speed limit
• c3c8102 Merge pull request #398 from hmaarrfk/lint
• a095aaa chore: add missing space for help of --output
• 10e8c85 lint other files
• eeb6995 Use .part suffix to represent partially downloaded files
• Additional commits viewable in compare view

Updates gdown from 4.7.3 to 5.2.2

Release notes

Sourced from gdown's releases.

v5.2.2

Security

• Fix path traversal vulnerability in extractall() that allowed zip/tar archives with ../ entries to write files outside the target directory (GHSA-76hw-p97h-883f)
• Reject symlinks, hardlinks, and special files in tar archives
• Use Python 3.12+ filter="data" for safe tar extraction when available
• Sanitize filenames from HTTP responses and URLs to prevent path traversal via /, , .., and null bytes
• Sanitize root folder name in download_folder() before building directory paths

v5.2.1

Fixes

• cached_download: Verify file hash before moving to final location instead of after (wkentaro/gdown#417)
  • Previously, the hash was verified after moving to the final path, which could leave corrupted files in place if hash verification failed
• download: Fix speed limit throttling logic to use independent byte counter instead of pbar.n (wkentaro/gdown#407)
  • The speed limiter now correctly tracks downloaded bytes independently from the progress bar
  • Also fixes unnecessary sleep when resuming downloads with speed limit (since pbar.n includes start_size from resumed downloads)

Chores

• Fix missing space in --output help text (wkentaro/gdown#398)
• Fix concatenated string literal in extractall.py error message

v5.2.0

🚀 Features

• Feature: preserve GDrive's last modified time when downloading #351 (Thanks @​moulins)
• Feature: allow folders to be continued more usefully #288 (Thanks @​achadwick)

✨ Enhancement

• Show resume progress bar from the middle #361

🐛 Fixes

• Use importlib.metadata to get the version number from the installed package #319
• BUG: Download folder with "skip_download=True" creates an output folder #321 (Thanks @​o-laurent)

💬 Other

• Fix typos in README.md #323 (Thanks @​amorehead)
• Run GitHub Actions on macOS, Ubuntu, and Windows #324
• Support Windows and macOS in tests #359
• Typo in arg help text #343 (Thanks @​frosit)
• Trivial: Update the documentation for the output param #347 (Thanks @​abouelkhair5)

---

• To connect, check my Twitter.
• To support, check Sponsoring Page.

... (truncated)

Commits

• af569fc fix: prevent path traversal in archive extraction and filename handling
• 7f4cb68 Merge pull request #417 from wkentaro/fix_cached_download_verify_file_hash_be...
• 9697a53 fix(cached_download): verify file hash before moving to final location
• c7c1b9d Merge pull request #407 from wkentaro/fix_speed_limit
• 940bd40 Fix throttling logic to use independent byte counter instead of pbar.n
• a55ce67 Fix unnecessary sleep when resuming downloads with speed limit
• c3c8102 Merge pull request #398 from hmaarrfk/lint
• a095aaa chore: add missing space for help of --output
• 10e8c85 lint other files
• eeb6995 Use .part suffix to represent partially downloaded files
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model generates dense vector embeddings for various tasks including search, clustering, and classification using specific instruction prefixes.

Links: Documentation | Paper

• Internalise the NomicBERT model (#43067) by @​ed22699 in #43067

MusicFlamingo

Music Flamingo is a fully open large audio–language model designed for robust understanding and reasoning over music. It builds upon the Audio Flamingo 3 architecture by including Rotary Time Embeddings (RoTE), which injects temporal position information to enable the model to handle audio sequences up to 20 minutes. The model features a unified audio encoder across speech, sound, and music with special sound boundary tokens for improved audio sequence modeling.

Links: Documentation | Paper

• Add Music Flamingo (#43538) by @​lashahub in #43538

... (truncated)

Commits

• c1c3424 update
• 20bff68 update release workflow
• 8956441 v5.5.0
• 5135e5e casually dropping the most capable open weights on the planet (#45192)
• a594e09 Internalise the NomicBERT model (#43067)
• 4932e97 Fix resized LM head weights being overwritten by post_init (#45079)
• 57e8413 [Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration (#45124)
• b10552e Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate ...
• 423f2a3 fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel (#...
• ade7a05 Generalize gemma vision mask to videos (#45185)
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model generates dense vector embeddings for various tasks including search, clustering, and classification using specific instruction prefixes.

Links: Documentation | Paper

• Internalise the NomicBERT model (#43067) by @​ed22699 in #43067

MusicFlamingo

Music Flamingo is a fully open large audio–language model designed for robust understanding and reasoning over music. It builds upon the Audio Flamingo 3 architecture by including Rotary Time Embeddings (RoTE), which injects temporal position information to enable the model to handle audio sequences up to 20 minutes. The model features a unified audio encoder across speech, sound, and music with special sound boundary tokens for improved audio sequence modeling.

Links: Documentation | Paper

• Add Music Flamingo (#43538) by @​lashahub in #43538

... (truncated)

Commits

• c1c3424 update
• 20bff68 update release workflow
• 8956441 v5.5.0
• 5135e5e casually dropping the most capable open weights on the planet (#45192)
• a594e09 Internalise the NomicBERT model (#43067)
• 4932e97 Fix resized LM head weights being overwritten by post_init (#45079)
• 57e8413 [Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration (#45124)
• b10552e Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate ...
• 423f2a3 fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel (#...
• ade7a05 Generalize gemma vision mask to videos (#45185)
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model generates dense vector embeddings for various tasks including search, clustering, and classification using specific instruction prefixes.

Links: Documentation | Paper

• Internalise the NomicBERT model (#43067) by @​ed22699 in #43067

MusicFlamingo

Music Flamingo is a fully open large audio–language model designed for robust understanding and reasoning over music. It builds upon the Audio Flamingo 3 architecture by including Rotary Time Embeddings (RoTE), which injects temporal position information to enable the model to handle audio sequences up to 20 minutes. The model features a unified audio encoder across speech, sound, and music with special sound boundary tokens for improved audio sequence modeling.

Links: Documentation | Paper

• Add Music Flamingo (#43538) by @​lashahub in #43538

... (truncated)

Commits

• c1c3424 update
• 20bff68 update release workflow
• 8956441 v5.5.0
• 5135e5e casually dropping the most capable open weights on the planet (#45192)
• a594e09 Internalise the NomicBERT model (#43067)
• 4932e97 Fix resized LM head weights being overwritten by post_init (#45079)
• 57e8413 [Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration (#45124)
• b10552e Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate ...
• 423f2a3 fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel (#...
• ade7a05 Generalize gemma vision mask to videos (#45185)
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model generates dense vector embeddings for various tasks including search, clustering, and classification using specific instruction prefixes.

Links: Documentation | Paper

• Internalise the NomicBERT model (#43067) by @​ed22699 in #43067

MusicFlamingo

Music Flamingo is a fully open large audio–language model designed for robust understanding and reasoning over music. It builds upon the Audio Flamingo 3 architecture by including Rotary Time Embeddings (RoTE), which injects temporal position information to enable the model to handle audio sequences up to 20 minutes. The model features a unified audio encoder across speech, sound, and music with special sound boundary tokens for improved audio sequence modeling.

Links: Documentation | Paper

• Add Music Flamingo (#43538) by @​lashahub in #43538

... (truncated)

Commits

• c1c3424 update
• 20bff68 update release workflow
• 8956441 v5.5.0
• 5135e5e casually dropping the most capable open weights on the planet (#45192)
• a594e09 Internalise the NomicBERT model (#43067)
• 4932e97 Fix resized LM head weights being overwritten by post_init (#45079)
• 57e8413 [Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration (#45124)
• b10552e Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate ...
• 423f2a3 fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel (#...
• ade7a05 Generalize gemma vision mask to videos (#45185)
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model generates dense vector embeddings for various tasks including search, clustering, and classification using specific instruction prefixes.

Links: Documentation | Paper

• Internalise the NomicBERT model (#43067) by @​ed22699 in #43067

MusicFlamingo

Music Flamingo is a fully open large audio–language model designed for robust understanding and reasoning over music. It builds upon the Audio Flamingo 3 architecture by including Rotary Time Embeddings (RoTE), which injects temporal position information to enable the model to handle audio sequences up to 20 minutes. The model features a unified audio encoder across speech, sound, and music with special sound boundary tokens for improved audio sequence modeling.

Links: Documentation | Paper

• Add Music Flamingo (#43538) by @​lashahub in #43538

... (truncated)

Commits

• c1c3424 update
• 20bff68 update release workflow
• 8956441 v5.5.0
• 5135e5e casually dropping the most capable open weights on the planet (#45192)
• a594e09 Internalise the NomicBERT model (#43067)
• 4932e97 Fix resized LM head weights being overwritten by post_init (#45079)
• 57e8413 [Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration (#45124)
• b10552e Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate ...
• 423f2a3 fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel (#...
• ade7a05 Generalize gemma vision mask to videos (#45185)
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model generates dense vector embeddings for various tasks including search, clustering, and classification using specific instruction prefixes.

Links: Documentation | Paper

• Internalise the NomicBERT model (#43067) by @​ed22699 in #43067

MusicFlamingo

Music Flamingo is a fully open large audio–language model designed for robust understanding and reasoning over music. It builds upon the Audio Flamingo 3 architecture by including Rotary Time Embeddings (RoTE), which injects temporal position information to enable the model to handle audio sequences up to 20 minutes. The model features a unified audio encoder across speech, sound, and music with special sound boundary tokens for improved audio sequence modeling.

Links: Documentation | Paper

• Add Music Flamingo (#43538) by @​lashahub in #43538

... (truncated)

Commits

• c1c3424 update
• 20bff68 update release workflow
• 8956441 v5.5.0
• 5135e5e casually dropping the most capable open weights on the planet (#45192)
• a594e09 Internalise the NomicBERT model (#43067)
• 4932e97 Fix resized LM head weights being overwritten by post_init (#45079)
• 57e8413 [Qwen3.5 MoE] Add _tp_plan to ForConditionalGeneration (#45124)
• b10552e Fix TypeError: 'NoneType' object is not iterable in GenerationMixin.generate ...
• 423f2a3 fix(models): Fix dtype mismatch in SwitchTransformers and TimmWrapperModel (#...
• ade7a05 Generalize gemma vision mask to videos (#45185)
• Additional commits viewable in compare view

Updates transformers from 4.57.3 to 5.5.0

Release notes

Sourced from transformers's releases.

Release v5.5.0

New Model additions

Gemma4

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

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

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

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

[!IMPORTANT] Gemma 4 does not apply the standard ImageNet mean/std normalization that many other vision models use. The model's own patch embedding layer handles the final scaling internally (shifting values to the [-1, 1] range).

The number of "soft tokens" (aka vision tokens) an image processor can produce is configurable. The supported options are outlined below and the default is 280 soft tokens per image.

Soft Tokens	Patches (before pooling)	Approx. Image Area
70	630	~161K pixels
140	1,260	~323K pixels
280	2,520	~645K pixels
560	5,040	~1.3M pixels
1,120	10,080	~2.6M pixels

To encode positional information for each patch in the image, Gemma 4 uses a learned 2D position embedding table. The position table stores up to 10,240 positions per axis, which allows the model to handle very large images. Each position is a learned vector of the same dimensions as the patch embedding. The 2D RoPE which Gemma 4 uses independently rotate half the attention head dimensions for the x-axis and the other half for the y-axis. This allows the model to understand spatial relationships like "above," "below," "left of," and "right of."

NomicBERT

NomicBERT is a BERT-inspired encoder model that applies Rotary Position Embeddings (RoPE) to create reproducible long context text embeddings. It is the first fully reproducible, open-source text embedding model with 8192 context length that outperforms both OpenAI Ada-002 and OpenAI text-embedding-3-small on short-context MTEB and long context LoCo benchmarks. The model gener...

Description has been truncated