llama.cpp b9723 adds Eagle3 speculation for Qwen

View on Reddit

llama.cpp's latest release (b9723) introduces Eagle3 speculative decoding support for Qwen models. Users enable it with `--spec-type draft-eagle3` and must provide a draft model.