Change8

v0.33.0

📦 huggingface-hubView on GitHub →
4 features🐛 11 fixes🔧 4 symbols

Summary

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.

✨ New Features

  • Added support for Featherless.ai as an inference provider, offering serverless access to a large catalog of models.
  • Added support for Groq as an inference provider, leveraging Language Processing Units (LPUs) for fast inference.
  • Enabled running tiny-agents using a local server endpoint, such as llama.cpp, moving towards 100% local agents.
  • Added local/remote endpoint inference support for MCP.

🐛 Bug Fixes

  • Fixed CLI exit issues in tiny-agents.
  • Re-injected JSON parse & runtime tool errors back into the chat history for MCP.
  • Made hf-xet more silent.
  • Fixed handling of collections within collections in HfApi.
  • Fixed inference search functionality.
  • Added warning if an inference provider's status is in error mode.
  • Fixed snapshot_download behavior on very large repositories (over 50k files).
  • Fixed tqdm_class argument when using subclasses of tqdm.
  • Fixed quality issues.
  • Corrected the second example in OAuth documentation.
  • Fixed table question answering.

Affected Symbols