CYBER WATCH / EQUINOXE INTELLIGENCE
← All articles
b12347 : llama : prune (#30565)
llama.cpp Releases · 2026-11-13 10:01:08
This update removes several deprecated APIs and fields, including llama_set_warmup, the defrag_thold field and the --defrag-thold option, as well as the obsolete llama-ext.h inclusion. It documents the llama_model_dflash_selector_top_k function and promotes the NextN/speculative APIs from llama-ext.h to llama.h, renaming llama_set_embd_state_src to llama_set_ctx_state_src and the internal equivalent. The parameters of these functions are renamed to ctx_src, NextN model queries are moved to the Adapters section, and redundant llama-ext.h includes in common/speculative.cpp and tools/imatrix are removed.Read original source ↗