v1.9.0.rc0
Breaking Changes📦 huggingface-hubView on GitHub →
⚠ 2 breaking✨ 6 features🐛 6 fixes🔧 5 symbols
Summary
This release introduces mounted volumes for Hugging Face Spaces, replacing deprecated persistent storage, and significantly enhances the CLI by adding agent-aware output formatting and support for installing marketplace skills.
⚠️ Breaking Changes
- The deprecated persistent storage feature is replaced by the new mounted volumes feature for Hugging Face Spaces. Users relying on persistent storage must migrate to using the new Volume API or CLI flags.
- The minimum required Python version has been bumped from 3.9 to 3.10.
Migration Steps
- If using persistent storage in Spaces, migrate configuration to use the new `Volume` API or CLI `--volume` / `-v` flags.
- Ensure your Python environment is running Python 3.10 or newer.
✨ New Features
- Hugging Face Spaces now support mounting volumes (models, datasets, buckets) directly via filesystem access using `set_space_volumes` or during creation/duplication.
- The `hf` CLI now auto-detects when running inside an AI coding agent environment and adapts output format (disabling ANSI, using TSV/compact JSON) for better machine readability.
- The `--format` flag can now explicitly force agent-friendly output (`agent`, `json`, `quiet`) in the CLI.
- The `hf skills add` command now supports installing skills directly from the Hugging Face skills marketplace.
- The CLI now auto-installs official HF CLI extensions on first invocation.
- Added `HF_HUB_DISABLE_SYMLINKS` environment variable to force cache operations to avoid using symlinks.
🐛 Bug Fixes
- Fixed PyTorchModelHubMixin failing to call .eval() when loading safetensors models.
- Validated shard filenames within sharded checkpoint index files.
- Fixed an issue in the `test_create_commit_conflict` test.
- Prevented scanning of CACHEDIR.TAG files within the cache.
- Deduplicated repository folder name generation logic.
- Rejected cross-type confusion between boolean and integer types during strict dataclass validation using @strict.