]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: cb8f4fa)
impl : use 6 digits for tensor dims (#20094)
authorddh0 <redacted>
Wed, 4 Mar 2026 08:53:38 +0000 (02:53 -0600)
committerGitHub <redacted>
Wed, 4 Mar 2026 08:53:38 +0000 (09:53 +0100)
commitc99909dd0b6589070e2a129fa172c40fa1675453
tree474bfe7308f9d5843398f8526a3e1d3520cc101ftree
parentcb8f4fa3f8e359e61ed570adb8a3c2ce74be5179commit | diff
impl : use 6 digits for tensor dims (#20094)

Many models have vocabulary sizes, and thus tensor shapes, with more
than 5 digits (ex: Gemma 3's vocab size is 262,208).

I already fixed this for `llama_format_tensor_shape` but missed it for
`llama_format_tensor_shape` until now. Oops.
src/llama-impl.cpp diff | blob | history
Packaging of ggml-org/llama.cpp