]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
overview / pkg / ggml / sources / llama.cpp / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 4937ca8)
model: rotate injected K/V cache for DFlash (#25823)
authorRuixiang Wang <redacted>
Sat, 18 Jul 2026 13:02:18 +0000 (15:02 +0200)
committerGitHub <redacted>
Sat, 18 Jul 2026 13:02:18 +0000 (15:02 +0200)
commit571d0d540df04f25298d0e159e520d9fc62ed121
tree1346c481f7a7c9ecfb514be667606d9253250cd3tree
parent4937ca83f4f3da63004943fe05d8aa4f0217d238commit | diff
model: rotate injected K/V cache for DFlash (#25823)

* dflash: rotate injected K/V cache when using K/V quantization

* Update src/models/dflash.cpp

Co-authored-by: Georgi Gerganov <redacted>
* clearer format

* remove trailing whitespace

---------

Co-authored-by: Georgi Gerganov <redacted>
src/models/dflash.cpp diff | blob | history
Packaging of ggml-org/llama.cpp