The vLLM project has released version 0.25.0rc2, addressing specific issues within the Transformers modelling backend.

  • Fixes embed scaling issues.
  • Resolves problems with CUDA graphs.