]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 8ea8fee)
llama-quant : default ftype param `Q5_1` --> `Q8_0` (#20828)
authorddh0 <redacted>
Sat, 25 Apr 2026 06:25:35 +0000 (01:25 -0500)
committerGitHub <redacted>
Sat, 25 Apr 2026 06:25:35 +0000 (09:25 +0300)
commit9d34231bb89590ee760ae19ba665e7855cd4fd4e
tree0e72fe6a5151bcfc19c2e47d2ee6761314b1cd01tree
parent8ea8fee966f50bd753ea1c916fbb723fa911999acommit | diff
llama-quant : default ftype param `Q5_1` --> `Q8_0` (#20828)

Change the default `ftype` in `llama_model_quantize_params` from
`LLAMA_FTYPE_MOSTLY_Q5_1` to `LLAMA_FTYPE_MOSTLY_Q8_0`.

In case some external program naively uses the default quantization
params, we should probably default to a known-good type like Q8_0 rather
than Q5_1, which is rather old.
src/llama-quant.cpp diff | blob | history
Packaging of ggml-org/llama.cpp