v0.35.0
📦 huggingface-hubView on GitHub →
✨ 8 features🐛 12 fixes⚡ 1 deprecations🔧 9 symbols
Summary
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.
Migration Steps
- If you rely on filtering models using library, tags, or task parameters in `list_models`, update your calls to use the new filtering mechanisms (if available, or adjust logic as filtering is deprecated).
✨ New Features
- Introduction of Scheduled Jobs via CLI, allowing cron-like execution of compute tasks on the Hugging Face Hub.
- New CLI commands for managing scheduled jobs: `hf jobs scheduled run`, `ps`, `inspect`, `delete`, `suspend`, and `resume`.
- Support for running commands using `uv run` within Jobs, including passing secrets like HF_TOKEN.
- Integration of Scaleway and PublicAI as new Inference Providers.
- Support for the Image-to-Video task in `InferenceClient`.
- Addition of a `.reasoning` field to Chat Completion output to capture reasoning tokens from certain providers.
- The `tiny-agents` feature now supports using `AGENTS.md` in addition to `PROMPT.md` for instructions.
- Added `apps` as a parameter to `HfApi.list_models`.
🐛 Bug Fixes
- Fixed incorrect total size calculation when resuming downloads.
- Fixed path extension logic to only apply on Windows systems.
- Improved Git Credential Helper Detection on Linux by adding GCM and libsecret support.
- Ensured requests decode content correctly.
- Added validation warnings for repository limits during `upload_large_folder`.
- Included `HF_HUB_DISABLE_XET` in the environment dump.
- Added type information for job owners.
- Switched to using only summary bars for uploads when running in notebooks.
- Improved error message for better shell compatibility.
- Fixed bash history expansion issues in HF Jobs examples documentation.
- Handled Ollama's deviation from the OpenAI tool streaming specification in MCP.
- Fixed tool allowance logic to correctly filter tools from MCP servers.
Affected Symbols
⚡ Deprecations
- Filtering by library, tags, or task in `list_models` is deprecated.