rendezqueue / rendezllama / 6725686114

90%

trunk: 87%

push

github

update(llama.cpp): with min-p sampling

This also includes a change to call llama_kv_cache_seq_rm()
instead of llama_kv_cache_tokens_rm(), which was removed upstream
because it only works for code tha calls the deprecated llama_eval().
We use llama_decode() instead of llama_eval(),
so all rewriting was probably a bit broken.

2 of 2 new or added lines in 1 file covered. (100.0%)

891 of 1081 relevant lines covered (82.42%)

14.09 hits per line