Qwen3.8 27B with embeddings

A discussion on reddit.com explores the Qwen3.8 27B model's built-in embedding support. Users are seeking ways to maintain normal generation performance in llama.cpp while embeddings are enabled. The conversation also touches upon the possibility of llama.cpp switching between different tasks without needing to unload and reload the model, indicating a desire for more efficient task handling with the Qwen3.8 27B model.

Unlike other models, Qwen3.8 27B's built-in embedding support raises new questions about optimizing performance within llama.cpp, moving beyond simple model integration.