The llama.cpp project has updated its server component to support vision input for the Clef model. This change is part of pull request #29969.

  • Added vision input support for Clef.
  • Moved input_attn_causal to private.
  • Extended old server_batch::embd.
  • Changed server_batch::token::pos to multi dim.
  • Fixed abort handling.
  • Fixed image token cap.
  • Fixed yield_to_queue mutate data issue.