Core NLP/ML library for state-of-the-art models
npx @buildinternet/releases get transformersAdds the Beam model from Reflection AI, a token-efficient frontier-scale model for coding and agentic use cases. Restores loading models on older torch versions, including from_pretrained on torch 2.7 and DtensorShardOperation on torch 2.6 and 2.7, plus fixes for HunYuan-VL image assignment during batched generation and DeepGEMM Mega MoE under expert parallelism.
Transformers v5.20.0 adds eight model architectures including Meta's 816B-parameter multimodal Muse Spark Open, InclusionAI's Ling 3.0 Flash hybrid MoE, Baidu's UnlimitedOcr, and MOSS-Transcribe-Diarize. It also removes deprecated cache utility methods and cross-model imports, raises the minimum kernels version to 0.17.2, and changes how LUKE and mLUKE tokenizers treat entity special tokens.
Fixes a breaking truncation and overflowing regression introduced in v0.23.1, and loosens the huggingface_hub requirement to allow v2, unblocking transformers installs. Also fixes type hints not resolving for pyright.
Adds support for EmbeddingGemma 2, an open multimodal embedding model that maps text, images, video, and audio into a single unified 768-dimensional vector space. Text search works via the feature-extraction pipeline, while other modalities use AutoProcessor and AutoModel directly, with all inputs comparable through cosine similarity.
MoE models that compute router logits now return them when output_router_logits=True, changing existing outputs, and the "paged|" prefix for SDPA and flash attention is deprecated in favor of setting the regular attention implementation. Adds EmbeddingGemma2, a multimodal embedding model handling text, images, audio, and video in a shared 768-dimensional space with Matryoshka truncation.
Datasets 5.1.0 adds the Vortex columnar format, Harbor RL environments, and biological formats including FASTA, FASTQ, GenBank, PDB, and mmCIF with BioSequence/BioStructure decoding. Fixes include an archive path check that allowed sibling-directory escapes, disallowed HDF5 external links, and an infinite-repetition bug in empty iterable datasets.
New model additions include Nemotron 3 Diarization (streaming speaker diarization, up to eight speakers), NemotronH Omni (joint text/image/video/audio reasoning), HyperCLOVAX Vision V2, and the GTE BERT-style encoder family. The release also carries flagged breaking changes across ROCm gpt-oss attention routing, DETR image processing, indexer layer_type remapping, vLLM video token counting, DINO modernization, and a Kernels version bump, plus a long tail of bug fixes spanning MoE expert parallelism, Qwen2.5-VL temporal RoPE, and DeepSeek-OCR-2 patching.
The tokenizers library has been refactored so its stack is faster than before, with a crate size 6x smaller and reduced peak memory, while keeping the same API, same token IDs, and same standards. The work focused on reducing the UTF-8 tax so improvements span across all languages.
Adds a dependency-free @huggingface/transformers-structured-output package that constrains generation to a JSON schema, JSON object, or regular expression, plus three new model architectures (DeepSeek-V4, Zaya, HRM-Text) and WebGPU support for Safari 26 and above. Fixes duplicate model file downloads when progress_callback is active, RawAudio.toBlob() handling of typed array byte offsets, and Moonshine ASR token decoding.
The final v0 release adds a Parity-aware BPE implementation with an example and trainer API docs entry, and avoids a full vocab clone in get_vocab_size(). Maintainers note this is the last v0 release before moving to v1.
Fixed DFlash candidate token device mismatch with device_map='auto', aligned logit distributions for sampling-based candidate generators, fixed MTP config when mlp_layer_types is absent, and added a fallback from Lanczos to bicubic filtering on CUDA so images process on accelerator. Also fixed gemma4 video device placement.
Adds support for Meta's Muse Glimmer 30B multimodal model plus GraniteSWA, GraniteMoeSWA, A.X-K1/K2, and Cosmos3 Edge. Kernels for linear attention models (Mamba, GDN, Conv-only) are now opt-in rather than mandatory, cache cropping only accepts negative offsets, and T5 family gains SDPA support with a possible default attention change.
Fixed a symlink-following arbitrary file write in archive extraction and a path traversal via metadata file_name in folder-based builders. Also fixed batch(by_column=...) crashing after shard/shuffle/split, CSV loader issues on pandas 2.0-2.2, and more than 20 other bugs across Arrow, JSON, and dataset card handling.
Fixed assisted decoding for models using EncoderDecoderCache (e.g., OlmoHybrid) and a SDPA prefill issue with position_bias during StaticCache. Also bumps FP8 kernels and fixes DeepGEMM on multiple devices.
Added Inkling, a 975B-parameter multimodal model, and TIPSv2. GPTNeoX now remaps embed_out to lm_head and GPTBigCode has attention backend changes for vLLM compatibility. Multi-Token Prediction decoding support, SDPA prefill with FlashAttention for StaticCache (up to 260% faster), and numerous bug fixes across MoE, cache, and generation.
Fixes custom model compatibility with the latest vllm release by being more defensive with remap_legacy_layer_types and handling cases where custom code doesn't know about the new linear layer type names. Also fixed a key type assertion in _LazyAutoMapping.register.