]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 04eb4c4)
kv-cache: follow the source cache size when sharing cells (#24267)
authorPascal <redacted>
Sun, 7 Jun 2026 15:33:00 +0000 (17:33 +0200)
committerGitHub <redacted>
Sun, 7 Jun 2026 15:33:00 +0000 (18:33 +0300)
commitf0156d1401500512ad85042ccf38970568b12253
tree9b6ccfd74c16e612902d1b07622146a162668185tree
parent04eb4c446d22b63449d5dc41c038987d4d8cc3a6commit | diff
kv-cache: follow the source cache size when sharing cells (#24267)

A fitted target context can end up smaller than the draft default, the
oversized assistant views then overflow the shared K/V tensors and trip
the ggml_view_4d size assert during graph reserve.
src/llama-kv-cache.cpp diff | blob | history
Packaging of ggml-org/llama.cpp