'text generate ' node too slow

llama-cpp plugin can run gguf 10 times faster than 'text generate ' node . can we use int4 text encoder to speed too？

https://github.com/lihaoyun6/ComfyUI-llama-cpp_vlm

Start by locating the “text generate” node and compare its GGUF and text-encoder handling with the linked ComfyUI-llama-cpp_vlm plugin. Benchmark the current path and determine whether int4 text-encoder support is feasible; done means a documented, reproducible speed improvement without changing generation behavior.

Written by the indexing model from the issue text.