Why not run Qwen 3.8 Flash Next with the full n-gram-table?

Already posted here https://www.reddit.com/r/LocalLLaMA/comments/1x85hfj/strata_qwen_38_flash_next_fast_but_whats_it/phavui9/ , but I believe this topic deserves it's own post. The PLE, aka the n-gram table that sits on the SSD, is quantized heavily in the common GGUFs and has a measurable impact o

中文全文翻译正在补齐，可先查看下方英文原文。

Already posted here https://www.reddit.com/r/LocalLLaMA/comments/1x85hfj/strata_qwen_38_flash_next_fast_but_whats_it/phavui9/ , but I believe this topic deserves it's own post. The PLE, aka the n-gram table that sits on the SSD, is quantized heavily in the common GGUFs and has a measurable impact o