Hugging Face

Browser-based OAuth Device Code flow is now the default login method, and hf upload uses a streamed multi-commit pipeline for faster, resumable large folder uploads. Also adds job wait commands, SSH access to running Jobs, and custom-container deployment for Inference Endpoints, with breaking changes around multi-commit uploads and non-detached job exit codes.

This release introduces the MiniMax-M3-VL vision-language model, the PP-OCRv6 OCR system, and the Parakeet-RNNT model for speech processing. Several bug fixes and improvements were also made, including changes to CI, stop string matching, and model documentation.

Enterprise organizations can now create service accounts: dedicated, organization-owned identities for programmatic access such as CI/CD pipelines...

The release introduces a new experimental A2POTrainer for optimal advantage regression and grants KTO trainer support for vision-language models. The AsyncRolloutWorker now runs in a separate process to avoid GIL contention and potential NCCL watchdog timeouts, along with fixes for aiohttp retries and all-NaN reward columns. Gold distillation trainer now aligns tokens via byte offsets, and SDFT/SDPO leverage the vLLM server for live teacher logprobs. Other features include bidirectional masked importance sampling for IcePop, support for NemotronH and Nemotron 3 Ultra, additional training chat templates, and decoupled self-distillation trainers.

Accelerate now works end-to-end on AMD ROCm devices. This release also includes numerous FSDP2 fixes and quality-of-life improvements, such as correct dtype handling on load, sharding of embeddings/norms, and QLoRA crash prevention.

Keyless CI/CD authentication is now supported via Trusted Publishers, allowing workflows to authenticate without storing an HF_TOKEN secret. The CLI now accepts hf:// URIs for uploading and downloading, and jobs can expose container ports through the public jobs proxy.

New models DiffusionGemma and DeepSeek-V3.2 have been added, featuring optimizations for inference speed and efficient long-context handling. The Kernels API was extended for module fusion and parameter transformation, with added support for fp8/fp4 Triton kernels. Model parallel beam search bugs in Qwen2-VL model families were fixed.

You can now publish to Hugging Face repositories from GitHub, GitLab, or custom CI systems without adding secrets, using workflow identity federation....

Agent traces from Claude Code, Pi, Codex and others can now be loaded with load_dataset and parsed to messages for training with TRL. Streaming shuffle now uses multiple input shards by default for better randomization, though this is a breaking change—use max_buffer_input_shards=1 to restore the prior behavior. Also added batch(by_column=...) for robotics datasets, support for Apache Iceberg, TsFile (IoTDB), 3D mesh, and CoNLL formats, plus fixes for Parquet streaming hangs, Lance file handling, and JSON null encoding.

Fixed a conversion bug for CLIP models that affected downstream models like SAM3.

Remote WasmExecutor has been removed. Exa is now available as a search engine option in WebSearchTool. Also includes documentation and typo cleanups, and a docstring update for LocalPythonExecutor.

You can now copy files or entire folders between different repositories without downloading or re-uploading, with LFS blobs deduplicated server-side. Added hf spaces ssh to open SSH sessions directly into a Space's Dev Mode container, and hf repos ls to list all your repositories with storage size and usage sorting. CLI tables now auto-fit to terminal width with per-column width caps and automatically right-align numeric columns.

Trainer telemetry is now gated on an explicit class-name allowlist, restricting which trainer classes can send telemetry.