HuggingFace Hub

This patch release fixes a critical bug related to file corruption during download retries when the server misbehaves regarding the Range header.

This patch release addresses a critical bug related to file corruption during download retries when the server misbehaves regarding the Range header.

This release introduces significant CLI enhancements, including a new `hf skills add` command for AI assistants and new commands for managing papers and collections. It also adds a decentralized evaluation results module with corresponding Python helpers and improves job execution support for multi-GPU training.

This release introduces significant enhancements to the CLI, including a new skill installation command (`hf skills add`), improved help output, and new commands for managing papers and collections. It also adds a decentralized evaluation results module to the Hub with corresponding Python helpers.

This patch release includes a minor bug fix related to logging HTTP error headers.

This release updates the default httpx timeout to use the HF_HUB_DOWNLOAD_TIMEOUT environment variable, providing flexibility for CI environments experiencing request timeouts.

This release fixes a minor regression in CommitInfo by setting the default value of _endpoint to None.

This release introduces the ability to list available hardware for Hugging Face Jobs through both CLI and API, alongside several bug fixes related to streaming performance and file system resolution. The maximum file size limit has also been increased.

This release introduces text-to-image support for the zai-org provider and fixes a bug related to endpoint forwarding in CommitUrl.

This release significantly reorganizes the CLI with dedicated discovery commands (`hf models`, `hf datasets`, `hf spaces`) and introduces real-time job monitoring via `hf jobs stats`. A breaking change removes the unused `direction` parameter from repository listing methods.

Version 1.2.4 fixes several bugs, including an issue with create_repo and job-related API endpoints, and adds the @dataclass_transform decorator to dataclass_with_extra.

This patch release fixes a bug in the CLI where new repository creation incorrectly defaulted to private=False instead of private=None, aligning CLI behavior with the API.

This patch release addresses two specific bugs: one related to reading corrupted metadata files and another concerning the display of the HF_TOKEN message in the authentication list.

This release significantly enhances rate limit handling with automatic retries and clearer errors, introduces the daily papers endpoint, and adds OVHcloud as a new Inference Provider. It also includes a breaking change where several access request listing functions now return iterators instead of lists.

This release introduces a convenient top-level import for HfFileSystem, now accessible as `huggingface_hub.hffs`.

This release focuses on stability, delivering several bug fixes across repository downloading, error handling, and CLI functionality.

This release introduces OVHcloud AI Endpoints as a new Inference Provider and significantly speeds up CLI installation by integrating `uv` support. Several minor bug fixes address issues in collections, CLI debugging, and inference parsing.

Version 1.1.4 introduces pagination for `list_user_access` results, which required a necessary breaking change to align with the updated server API.

This patch release introduces an optional 'name' parameter for catalog deployment and resolves rate-limiting issues encountered during large dataset downloads.

This release focuses on optimizing the file download experience through multi-threading and cleaner CLI output, while significantly expanding CLI capabilities with new commands for managing Inference Endpoints and verifying cache integrity. New support for WaveSpeedAI and image segmentation on fal are also introduced.

This patch release removes the lingering dependency on `aiohttp` from the `huggingface_hub[inference]` extra and cleans up an unused internal method.

Hugging Face Hub library v1.0 introduces a major migration to use httpx for HTTP requests, dropping requests/aiohttp dependencies, and completely revamps the CLI experience, replacing `huggingface-cli` with the new `hf` command.

This release focuses on significant performance optimizations for HfFileSystem and introduces the new get_organization_overview API endpoint, serving as the final minor release before v1.0.0.

This release addresses two specific bug fixes related to fal-ai image-to-image inference and Tiny-Agents tool usage.

This release introduces Z.ai as a new official Inference Provider on the Hub and includes an optimization for file system glob operations.

This patch release adjusts the retry logic to only target 5xx errors and fixes an issue with unresolved forward references in strict dataclasses.

This release introduces Scheduled Jobs for running compute tasks on a regular basis via a new CLI interface, alongside significant updates to the Inference Client supporting Image-to-Video and new providers. Several bug fixes and maintenance updates were also applied across the CLI and Jobs API.

This release introduces PublicAI as a new supported inference provider, allowing users to interact with models hosted by this nonprofit organization via InferenceClient.

This release introduces Scaleway as a new supported inference provider, allowing users to access models hosted on their infrastructure via the InferenceClient.

This release introduces support for the Image-To-Video inference task using Fal AI and includes several quality-of-life improvements across job handling, environment dumping, and repository uploads.

This release updates the uv image for Jobs and improves the whoami command output by adding a 'user:' prefix.

This patch release addresses two specific bugs: one related to path extension on Windows and another concerning incorrect total size reporting during download resumption.

Version 0.34.1 primarily addresses stability issues in file downloading utilities, fixing potential failures with private tokens and race conditions during concurrent downloads.

This release introduces 'Jobs', a powerful new CLI and Python API for running compute tasks on Hugging Face infrastructure using Docker-like commands. Additionally, the Hugging Face CLI has been officially renamed from `huggingface-cli` to `hf`, and the InferenceClient gained image-to-image support.

This patch release addresses a specific UserWarning related to open sessions when using AsyncInferenceClient for streaming operations.

This release introduces an update to tiny-agent by omitting parameters in its default tools. More details are available in the linked pull request.

This release primarily updates the tiny-agents example and provides a link to the full comparison between versions.

This release updates the Tiny-Agent integration to use the VSCode MCP format, requiring significant changes to configuration structure, including flattening nested mappings and moving request headers to the root level.

This patch release (v0.33.1) focuses on bug fixes related to payload preparation and inference endpoint health checks, alongside adding tool call support to Tiny agents messages.

This release introduces two major new inference providers, Featherless.AI and Groq, significantly enhancing model accessibility and inference speed. It also brings several quality-of-life improvements, bug fixes across various components, and advancements for local agent execution via MCP and tiny-agents.

This patch release fixes a bug related to the incorrect saving of upload_mode and remote_oid parameters.

This minor release introduces the ability to inject environment variables into headers and enhances the codebase with better type annotations.

This release focuses on bug fixes, specifically addressing issues in `asyncio` usage, token yielding, and the `InferenceClient.question_answering` method.

This release focuses on improvements and bug fixes for the tiny-agents feature, including better environment variable handling and CLI stability.

Version 0.32.2 introduces local/remote endpoint inference support and resolves a critical bug affecting snapshot downloads for extremely large repositories.

This is a patch release focused on fixing a specific reported issue (#3116).

This release introduces powerful new capabilities for LLM interaction via the Model Context Protocol (MCP) Client and Tiny Agents CLI, alongside support for new inference providers and enhanced dataclass validation with the @strict decorator.

This release introduces new `strict` decorators for dataclass validation and adds DTensor support to storage size helpers, alongside several bug fixes.

This patch release focuses on making the `hf-xet` dependency optional, improving installation flexibility. More context is available in related pull requests.

This release introduces major enhancements to Inference Providers, adding support for LoRA inference via fal.ai and Replicate, and enabling 'auto' provider selection as the new default. Additionally, Xet uploads now support byte arrays, and large file downloads (>50GB) are more reliable.

This patch release addresses several bugs related to the InferenceClient, specifically fixing text-generation with external providers and conversational handling in HfInference.

This patch release addresses a specific bug reported in the Hugging Face Hub repository.

This major release introduces Xet, a new protocol for storing large objects in Git, and significantly enhances the InferenceClient with new providers (Cerebras, Cohere, Novita) and organizational billing support. New features also include wildcard support for CLI uploads and programmatic LFS file management.

This release introduces client-side support for the Cerebras and Cohere providers in preparation for their official launch on the Hub.

This patch release (v0.29.2) addresses two specific bugs concerning payload model naming when using URLs and ensuring sys.stdout is restored after errors in notebook_login.

This patch release (v0.29.1) addresses two specific bugs related to large folder uploads and the inference endpoint waiting mechanism.

This release significantly expands serverless inference capabilities by adding official support for Fireworks AI, Hyperbolic, Nebius AI Studio, and Novita providers. It also includes several quality-of-life improvements and deprecates some legacy hf-inference specific features.

This patch release fixes a critical bug introduced in v0.28.0 related to setting the HF_ENDPOINT environment variable with subpaths.

This release introduces major updates to the InferenceClient, enabling unified inference across multiple third-party providers using Hugging Face Hub model IDs. Additionally, HfApi has been updated with new repository properties and the deprecated `like` endpoint has been removed.