llama.cpp build b12347 removes old features and promotes new ones
A routine cleanup of the llama.cpp local AI engine deletes retired code, including one command-line option, and moves newer features into the main interface.
The short version
- Removes the deprecated llama_set_warmup function and other APIs marked DEPRECATED in llama.h
- Drops the deprecated defrag_thold field and the --defrag-thold command-line option
- Moves the NextN and speculative API from llama-ext.h into llama.h
- Renames llama_set_embd_state_src to llama_set_ctx_state_src
- Each change credits DeepSeek-V4.1-Flash as an assistant, run on a Mac Studio
Why it matters
If you only run models with a recent build, you will likely see no difference. If you write software on top of llama.cpp or use scripts that pass the --defrag-thold option, this build can break them. It is a reminder to read change notes before swapping in a new version.
What the cleanup removes
As github.com reported in the notes for build b12347, the change titled 'llama : prune' is a bundle of small commits. The first deletes llama_set_warmup, a function that had already been marked deprecated, meaning flagged for eventual removal.
A second commit goes further. It drops every remaining interface marked DEPRECATED in the library's main header file, llama.h. It also removes a deprecated context setting called defrag_thold and the matching --defrag-thold command-line option.
What gets promoted
The other half of the work moves things up rather than out. A set of functions covering NextN, layer inputs, and embedding and state rows was kept in a secondary header, llama-ext.h. They now live in llama.h, the library's main public interface.
One function was renamed along the way, from llama_set_embd_state_src to llama_set_ctx_state_src. A follow-up commit tidied parameter names, moved the NextN draft-model queries next to the Adapters section, and removed includes that were no longer needed. Another commit added documentation for llama_model_dflash_selector_top_k.
An AI-assisted commit trail
Every commit carries an 'Assisted-by' line naming DeepSeek-V4.1-Flash-1017-GGUF, with a Mac Studio (M5 Ultra, 256GB) listed as the machine. The source does not say how much of the work the model did, so we will not guess.
Who should care
Developers who build on llama.cpp should check whether they call any removed function or the old name of the renamed one. Anyone with a launch script that passes --defrag-thold should remove it before updating.
Everyone else can treat this as housekeeping. Keep your last working build on hand in case an update misbehaves.
What to watch
- Whether downstream projects report breakage from the removed deprecated functions
- Whether the newly promoted NextN and speculative API gets documentation or examples
Try it
Search your launch scripts for --defrag-thold and remove it before updating to b12347 or later. Keep your previous build so you can roll back.
Sources and links
- Reportinggithub.com: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by github.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.