●Change8

v1.2.0

Breaking Changes
📦 huggingface-hubView on GitHub →
⚠ 1 breaking✨ 8 features🐛 6 fixes🔧 13 symbols

Summary

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.

⚠️ Breaking Changes

  • Paginated results from `list_pending_access_requests`, `list_accepted_access_requests`, and `list_rejected_access_requests` now return an iterator instead of a list. To maintain previous behavior, wrap the call with `list(...)`.

Migration Steps

  1. If you rely on the return type of `list_pending_access_requests`, `list_accepted_access_requests`, or `list_rejected_access_requests` being a list, wrap the call with `list(...)` to maintain the old behavior: e.g., `list(list_pending_access_requests(...))`.
  2. Use `huggingface_hub.is_offline_mode()` instead of checking `HF_HUB_OFFLINE` directly.
  3. When using `InferenceClient.feature_extraction`, ensure `truncation_direction` parameters use lowercase values ("left"/"right").

✨ New Features

  • Improved rate limit handling: SDK now automatically parses `RateLimit` headers during 429 errors, waits the specified duration, and provides clearer error messages showing retry time and remaining requests.
  • Added `list_daily_papers` endpoint to programmatically access Hugging Face's daily papers feed with filtering and sorting capabilities.
  • Introduced `huggingface_hub.is_offline_mode()` as the recommended utility to check if offline mode is enabled, replacing direct checks of `HF_HUB_OFFLINE`.
  • Inference Endpoints now support configuring scaling metrics and thresholds during deployment.
  • Exposed `RepoFile` and `RepoFolder` at the root level for easier imports.
  • Added OVHcloud AI Endpoints as an official Inference Provider.
  • Added support for Automatic Speech Recognition (ASR) via the Replicate Inference Provider.
  • A new top-level alias `hffs` is available for convenient access to the `HfFileSystem` interface.

🐛 Bug Fixes

  • Fixed zero-shot classification output parsing in Inference.
  • Resolved `FileNotFoundError` during CLI update checks.
  • Fixed `HfHubHTTPError` reduction error by adding a factory function.
  • Made 'name' optional in catalog deploy.
  • Prevented use of `rich` in the tiny-agents CLI.
  • Used `constants.HF_HUB_ETAG_TIMEOUT` as the timeout for `get_hf_file_metadata`.

Affected Symbols

list_pending_access_requestslist_accepted_access_requestslist_rejected_access_requestsHfHubHTTPErrorlist_daily_papersis_offline_modeHF_HUB_OFFLINEInferenceClient.feature_extractionRepoFileRepoFolderhffsInferenceClient.automatic_speech_recognitionget_hf_file_metadata