]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
common: resolve draft repo to its requested sidecar (#25955)
authorPascal <redacted>
Tue, 21 Jul 2026 16:03:43 +0000 (18:03 +0200)
committerGitHub <redacted>
Tue, 21 Jul 2026 16:03:43 +0000 (18:03 +0200)
commit60f6a17704163e0273bfadb9abb30deb14270f7f
treea65231971f2eb34326d673ea110e856c655ce1a4
parentfd41bf65a2e0013b1ecafa832f8ce6f18b1753e8
common: resolve draft repo to its requested sidecar (#25955)

With -hfd pointing to a repo shipping speculative sidecars, the draft
resolved to the main model of that repo, since find_best_model()
excludes sidecar files, and the explicit draft plan suppressed the
sidecar discovery on the -hf repo.

The draft plan already discovers its sidecars, they were just never
consumed. Wire them as the draft, following the fallback pattern of
the main plan, so this now works as expected:

llama-server -hf repo -hfd repo --spec-type draft-dflash
common/arg.cpp