v0.34.0
📦 huggingface-hubView on GitHub →
✨ 13 features🐛 12 fixes⚡ 1 deprecations🔧 5 symbols
Summary
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.
Migration Steps
- If using the CLI, update scripts and habits to use `hf` instead of `huggingface-cli` (e.g., use `hf jobs run` instead of `huggingface-cli jobs run`).
- If using Tiny-Agents, update configuration files to the new VSCode-like format to properly handle secrets and environment variables.
✨ New Features
- Introduction of the new `hf jobs` command-line interface for running and managing compute jobs on Hugging Face infrastructure.
- Jobs CLI supports familiar Docker-like commands: `run`, `ps`, `logs`, `inspect`, `cancel`.
- Jobs support running on various hardware flavors including CPUs, T4/A10G/A100 GPUs, and TPUs via a simple flag.
- Jobs can execute commands within Docker images, HF Spaces, or custom containers.
- Live log streaming for running jobs.
- Python API for jobs introduced with functions like `run_job` and `list_jobs`.
- Experimental UV Runner (`hf jobs uv run`) for running Python scripts with inline dependencies.
- The Hugging Face CLI has been officially renamed from `huggingface-cli` to `hf`.
- InferenceClient now supports `image-to-image` task for Replicate and fal.ai providers.
- InferenceClient now accepts `PIL.Image` objects directly as input.
- Tiny-Agents updated input format to align more closely with VSCode config format, improving environment variable and secret handling.
- Xet integration is now stable and production-ready, handling a majority of file transfers on new repos.
- Improved progress reporting for Xet uploads.
🐛 Bug Fixes
- Fixed headers handling and secrets management in Tiny-Agents.
- Kept empty dictionaries out of the chat completion payload in Tiny-Agents.
- Fixed recursive filtering in Inference Providers.
- Removed default parameter values for text generation in Inference.
- Correctly built chat completion URL with query parameters in Inference.
- Updated Tiny-Agents example.
- Fixed "failed to parse tools" error caused by mcp EXIT_LOOP_TOOLS not conforming to ChatCompletionInputFunctionDefinition model.
- Added tool call to messages in Tiny-Agents.
- Omitted parameters for default tools in Tiny-Agents.
- Fixed payload preparation to correctly merge parameters into the output dictionary.
- Fixed inference_endpoints health check to use GET `healthRoute` instead of GET /.
- Updated hf_api.py.
Affected Symbols
⚡ Deprecations
- The legacy `huggingface-cli` command is officially deprecated, although it remains available without breaking changes for now.