Change8

v4.57.4

📦 transformersView on GitHub →
🐛 3 fixes🔧 2 symbols

Summary

This patch release for v4 includes minor fixes for remote generation methods like grouped beam search, vLLM integration, and an offline tokenizer loading issue for Mistral models.

🐛 Bug Fixes

  • Fixed an issue where grouped beam search was incorrectly using config parameters.
  • Improved handling of decorators with optional arguments.
  • Fixed an issue preventing Mistral base model loading when offline if the tokenizer was already cached.

Affected Symbols