v4.51.3-Qwen2.5-Omni-preview is a pre-release release. Key changes include:

• A new model is added to transformers: Qwen2.5-Omni.;
• In order to install this version, please install with the following command:.

v4.51.3-InternVL-preview is a pre-release release. Key changes include:

• A new model is added to transformers: InternVL (2.5 & 3) It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-Intern…;
• In order to install this version, please install with the following command:.

v4.51.3-Janus-preview is a pre-release release. Key changes include:

• A new model is added to transformers: Janus It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-Janus-preview.;
• In order to install this version, please install with the following command:.

v4.51.3-TimesFM-preview is a pre-release release. Key changes include:

• A new model is added to transformers: TimesFM It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-TimesFM-preview.;
• In order to install this version, please install with the following command:.

v4.51.3-MLCD-preview is a pre-release release. Key changes include:

• A new model is added to transformers: MLCD It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-MLCD-preview.;
• In order to install this version, please install with the following command:.

v4.51.3 is a patch release. Key changes include:

• Handle torch ver in flexattn;
• handle torch version edge cases.

v4.51.2 is a patch release. Key changes include:

• Fix Llama4 offset (#37414) by @Cyrilvallez;
• Attention Quantization with FBGemm & TP (#37384) by @MekkCyber.

v4.51.1 is a patch release. Key changes include:

• Fixing flex attention for torch=2.6.0;
• more fixes for post-training llama4.

v4.51.0 is a minor release. Key changes include:

• The highly capable Llama 4 Maverick with 17B active parameters out of ~400B total, with 128 experts.;
• The efficient Llama 4 Scout also has 17B active parameters out of ~109B total, using just 16 experts..

v4.50.3-DeepSeek-3 is a pre-release release. Key changes include:

• current implementation uses the "naive" attention compution (so not really MLA);
• current implementation loops through the experts. This should be replaced. Pointers to use get_packed_weights from intetrations/tensor_parallel..

v4.50.3 is a patch release. Key changes include:

• [generate] beam search -- fix output cropping (#37080) by @gante;
• [blip-2] Fix dtype mismatch when keep in fp32 (#37068) by @zucchini-nlp.

v4.50.2 is a patch release. Key changes include:

• [Utils] torch version checks optionally accept dev versions (#36847) by @gante;
• Fix processor kwargs qwen2 vl (#36890) by @yonigozlan.

v4.50.1 is a patch release. Key changes include:

• Deprecate #36741 and map Causal to Conditional (#36917) by @zucchini-nlp;
• Fix pytorch deform attn path (#36923) by @qubvel.

v4.50.0 is a minor release. Key changes include:

• v4.49.0-Gemma-3;
• v4.49.0-AyaVision.

v4.49.0-Mistral-3 is a pre-release release. Key changes include:

• Fast-response conversational agents.;
• Low-latency function calling..

v4.49.0-Gemma-3 is a pre-release release. Key changes include:

• For image+text and image-only inputs use Gemma3ForConditionalGeneration.;
• For text-only inputs use Gemma3ForCausalLM for generation to avoid loading the vision tower..

v4.49.0-AyaVision is a pre-release release. Key changes include:

• Multimodal capabilities in 23 languages;
• Strong text-only multilingual capabilities inherited from CommandR-7B post-trained with the Aya Expanse recipe and Aya Expanse 32B.

v4.49.0-SigLIP-2 is a pre-release release. Key changes include:

• Usage of SigLIP2 is similar to SigLIP and CLIP. The main difference from CLIP is the training loss, which does not require a global view of all the pairwise si…;
• Training is supported but does not use torch.distributed utilities which may limit the scalability of batch size. However, DDP and FDSP works on single-node mu….

v4.49.0-SmolVLM-2 is a pre-release release. Key changes include:

• It uses SmolLM2 for the text model.;
• It supports multi-image and video inputs.

v4.49.0 is a minor release. Key changes include:

• Add-helium by @ArthurZucker in #35669;
• add qwen2.5vl by @ShuaiBai623 in #35569.

v4.48.3 is a patch release. Key changes include:

• Add future import for Py < 3.10 (#35666) by @Rocketknight1;
• Fix device in rope module when using dynamic updates (#35608) by @Cyrilvallez.

v4.48.2 is a patch release. Key changes include:

• Restore is_torch_greater_or_equal_than for backward compatibility (#35734) by @tlrmchlsmth;
• Fix NoneType type as it requires py>=3.10 (#35843) by @SunMarc.

v4.48.1 is a patch release. Key changes include:

• [Phi] bias should be True (#35650) @ArthurZucker;
• Fix condition when GA loss bug fix is not performed (#35651) @techkang.

v4.48.0 is a minor release. Key changes include:

• Rotary Positional Embeddings to support sequences of up to 8192 tokens.;
• Unpadding to ensure no compute is wasted on padding tokens, speeding up processing time for batches with mixed-length sequences..

v4.51.3-Qwen2.5-Omni-preview is a pre-release release. Key changes include:

• A new model is added to transformers: Qwen2.5-Omni.;
• In order to install this version, please install with the following command:.

v4.51.3-InternVL-preview is a pre-release release. Key changes include:

• A new model is added to transformers: InternVL (2.5 & 3) It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-Intern…;
• In order to install this version, please install with the following command:.

v4.51.3-Janus-preview is a pre-release release. Key changes include:

• A new model is added to transformers: Janus It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-Janus-preview.;
• In order to install this version, please install with the following command:.

v4.51.3-TimesFM-preview is a pre-release release. Key changes include:

• A new model is added to transformers: TimesFM It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-TimesFM-preview.;
• In order to install this version, please install with the following command:.

v4.51.3-MLCD-preview is a pre-release release. Key changes include:

• A new model is added to transformers: MLCD It is added on top of the v4.51.3 release, and can be installed from the following tag: v4.51.3-MLCD-preview.;
• In order to install this version, please install with the following command:.

v4.51.3 is a patch release. Key changes include:

• Handle torch ver in flexattn;
• handle torch version edge cases.

v4.51.2 is a patch release. Key changes include:

• Fix Llama4 offset (#37414) by @Cyrilvallez;
• Attention Quantization with FBGemm & TP (#37384) by @MekkCyber.

v4.51.1 is a patch release. Key changes include:

• Fixing flex attention for torch=2.6.0;
• more fixes for post-training llama4.

v4.51.0 is a minor release. Key changes include:

• The highly capable Llama 4 Maverick with 17B active parameters out of ~400B total, with 128 experts.;
• The efficient Llama 4 Scout also has 17B active parameters out of ~109B total, using just 16 experts..

v4.50.3-DeepSeek-3 is a pre-release release. Key changes include:

• current implementation uses the "naive" attention compution (so not really MLA);
• current implementation loops through the experts. This should be replaced. Pointers to use get_packed_weights from intetrations/tensor_parallel..

v4.50.3 is a patch release. Key changes include:

• [generate] beam search -- fix output cropping (#37080) by @gante;
• [blip-2] Fix dtype mismatch when keep in fp32 (#37068) by @zucchini-nlp.

v4.50.2 is a patch release. Key changes include:

• [Utils] torch version checks optionally accept dev versions (#36847) by @gante;
• Fix processor kwargs qwen2 vl (#36890) by @yonigozlan.

v4.50.1 is a patch release. Key changes include:

• Deprecate #36741 and map Causal to Conditional (#36917) by @zucchini-nlp;
• Fix pytorch deform attn path (#36923) by @qubvel.

v4.50.0 is a minor release. Key changes include:

• v4.49.0-Gemma-3;
• v4.49.0-AyaVision.

v4.49.0-Mistral-3 is a pre-release release. Key changes include:

• Fast-response conversational agents.;
• Low-latency function calling..

v4.49.0-Gemma-3 is a pre-release release. Key changes include:

• For image+text and image-only inputs use Gemma3ForConditionalGeneration.;
• For text-only inputs use Gemma3ForCausalLM for generation to avoid loading the vision tower..

v4.49.0-AyaVision is a pre-release release. Key changes include:

• Multimodal capabilities in 23 languages;
• Strong text-only multilingual capabilities inherited from CommandR-7B post-trained with the Aya Expanse recipe and Aya Expanse 32B.

v4.49.0-SigLIP-2 is a pre-release release. Key changes include:

• Usage of SigLIP2 is similar to SigLIP and CLIP. The main difference from CLIP is the training loss, which does not require a global view of all the pairwise si…;
• Training is supported but does not use torch.distributed utilities which may limit the scalability of batch size. However, DDP and FDSP works on single-node mu….

v4.49.0-SmolVLM-2 is a pre-release release. Key changes include:

• It uses SmolLM2 for the text model.;
• It supports multi-image and video inputs.

v4.49.0 is a minor release. Key changes include:

• Add-helium by @ArthurZucker in #35669;
• add qwen2.5vl by @ShuaiBai623 in #35569.

v4.48.3 is a patch release. Key changes include:

• Add future import for Py < 3.10 (#35666) by @Rocketknight1;
• Fix device in rope module when using dynamic updates (#35608) by @Cyrilvallez.

v4.48.2 is a patch release. Key changes include:

• Restore is_torch_greater_or_equal_than for backward compatibility (#35734) by @tlrmchlsmth;
• Fix NoneType type as it requires py>=3.10 (#35843) by @SunMarc.

v4.48.1 is a patch release. Key changes include:

• [Phi] bias should be True (#35650) @ArthurZucker;
• Fix condition when GA loss bug fix is not performed (#35651) @techkang.

v4.48.0 is a minor release. Key changes include:

• Rotary Positional Embeddings to support sequences of up to 8192 tokens.;
• Unpadding to ensure no compute is wasted on padding tokens, speeding up processing time for batches with mixed-length sequences..