]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: e2bd725)
llama : fix typical sampling (#4261)
authortarcey <redacted>
Thu, 30 Nov 2023 21:40:23 +0000 (22:40 +0100)
committerGitHub <redacted>
Thu, 30 Nov 2023 21:40:23 +0000 (23:40 +0200)
commit954e22858c5cea1dc03e9172d3879402af2b5990
treec8c80ab70b8953e70770c06ac1af4301d7cf431etree
parente2bd725f4b39bc5c6234858d158e01248f5ab5bdcommit | diff
llama : fix typical sampling (#4261)

Typical sampling was broken because after copying new_candidates into canditates, the "sorted" bool is left at "true", but the new data is no longer sorted according to probability. Patch to set "sorted" to false.

Test: Generating with temp=0.0001 (approx. argmax)  should generate the same sequence at typical>=1.0 and typical=0.9999 (approx. disabled, but enters the typical sampling codepath).
llama.cpp diff | blob | history
Packaging of ggml-org/llama.cpp