]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 6df686b)
convert : use n_groups instead of hardcoded values in reshape (#18929)
authorDaniel Bevenius <redacted>
Tue, 20 Jan 2026 05:55:24 +0000 (06:55 +0100)
committerGitHub <redacted>
Tue, 20 Jan 2026 05:55:24 +0000 (06:55 +0100)
commit7dee9ff59ad507304bf43a2682dbe0a89bbc3dce
tree730cb7e3e0792fc35c32487b4d6df389b18fed6dtree
parent6df686bee68ff109f62123c7a8eac003f3dd9e20commit | diff
convert : use n_groups instead of hardcoded values in reshape (#18929)

* convert : use n_groups instead of hardcoded values in reshape

This commit modifies the conversion script for NemotronHModel to use
the 'n_groups' hyperparameter, and allow Python to calculate the the
last dimension, using -1, when reshaping the 'mixer.norm.weight' tensor.

* use self.n_group instead of self.hparams["n_groups"]
convert_hf_to_gguf.py diff | blob | history
Packaging of ggml-org/llama.cpp