npx @buildinternet/releases get hugging-faceRecently shipped Transformers v5.19 with EmbeddingGemma2, Diffusers 0.41 with Qwen-Image 2.1, and Datasets 5.1's new data formats.
Transformers added a cross-modal embedder and streaming diarization.
Diffusers 0.41 shipped Qwen-Image 2.1. One model for generation and editing: a 7B visual component, multiple reference images, native RGBA transparency, and LoRA training. Minor releases now cluster around new model integrations; patches stay fix-only.
Datasets 5.1 widened format support. Vortex backs datasets with high-performance columnar files, Harbor covers RL environments, and FASTA, FASTQ, and GenBank handle DNA, protein, and molecular structure.
huggingface_hub 2.1 improved Jobs. Retry a failed Job with attempts, rerun one from its saved spec, reschedule a cron without recreating it, and expose ports with token-protected or public access. ZeroGPU quota tracking and a revamped Model Catalog landed too. The v2.0 breaking release stands: the HTTP stack moved to httpx2, so custom client factories and transport except clauses need httpx2 types.
TRL removed its vendored loss module. DPO, KTO, and GRPO now stream their own log-probs; the fused copies had drifted on GRPO grad_norm and KTO's reference model under PEFT, and never all-reduced gradients under plain DDP.
Tokenizers hit 1.0 RC with the same API and token IDs, a 6x smaller crate, lower peak memory, and faster tokenization across languages.
Smaller surfaces. PEFT 0.21 added Riemannian-preconditioned LoRA and KaSA. TEI v1.9.4 added ARM64 and multi-arch CUDA images plus minimal AMD ROCm support. timm v1.0.30 added Qwen-Drive vision tower weights and Qwen3-VL, Sapiens2, and iFormer variants.
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.
A new fused LM head, on by default in SFT, DPO, KTO, GRPO, RLOO and Distillation, projects hidden states through the LM head in Triton-kernel tiles to avoid building the full logits tensor, raising max trainable sequence length up to 6.9x and cutting peak memory 52% to 82%. The full-logits scoring path, use_liger_kernel chunked path and _forward_redirection are removed, so a PEFT adapter on lm_head now raises and use_liger_kernel=True is deprecated for v2.0.0 removal.
hf jobs stats and HfApi.fetch_job_metrics now return a single snapshot instead of streaming until job completion, with -f/--follow on the CLI and follow=True in Python restoring the previous behavior. Also fixed a race in _get_version that made concurrent downloads skip hf_xet, and multiple HfFileSystem cache issues including duplicate listings and stale not-found results.
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.
GRPO, RLOO, and Distillation on the default non-vLLM generation path now stop completions at every eos id the model declares, fixing silent training on tokens generated past the end of the turn that showed up as completions/clipped_ratio near 1. Also fixes DFT loss on batches with no trainable tokens, precompute_ref_log_probs under FSDP with no reference model, completion offsets where the tokenized prompt and prompt+completion diverge, and guards the requests and urllib3 imports in the vLLM client.
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.
Qwen-Image 2.1 unifies image generation and editing in one model with 7B visual generation parameters, multiple reference images, native RGBA transparency output, and LoRA training. ONNX support is deprecated in favor of Optimum, and previously deprecated LuminaText2ImgPipeline and Lumina2Text2ImgPipeline aliases were removed in favor of LuminaPipeline and Lumina2Pipeline.
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.
Users can now add a P(doom) estimate, their probability of AI causing an existential catastrophe, to their Hugging Face profile. Answers also feed a community survey on AI risk perception, with only aggregated, anonymized results published in two weeks.
Fixes an issue that prevented encoder-decoder models from working when using Transformers 5.18.0 or later.
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 trl.losses module introduced in v1.13 is removed, with DPO, KTO, and GRPO now streaming their own per-token log-probs through _ChunkedLogProbFunction, which also unlocks DPO with precomputed reference log-probs that did not exist before. A fused Triton logprob and entropy kernel now ships in-tree at trl.kernels and is on by default on CUDA, ROCm, and XPU, and six experimental trainers were removed from trl.experimental.
The job page has a new Resources panel next to the logs that shows live CPU, RAM, GPU, and VRAM usage while a job runs, updated every couple of seconds.
Organizations can now subscribe to Hugging Face through their Google Cloud account, putting Inference Providers, Inference Endpoints, Spaces, Jobs, storage, and Team or Enterprise seats on their existing cloud invoice. Subscription works like the AWS Marketplace integration: subscribe from the listing, pick the organization to link, and usage moves to the cloud bill once the subscription is validated.