]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
common : auto-download dflash- and eagle3- HF sidecars (#25811)
authorGeorgi Gerganov <redacted>
Fri, 17 Jul 2026 09:15:30 +0000 (12:15 +0300)
committerGitHub <redacted>
Fri, 17 Jul 2026 09:15:30 +0000 (12:15 +0300)
commit635cdd5fcc5bdeb8ec2e108bb2a40acf62d9039b
tree5939048a00cf16003a27cf538ab010a0c79de50a
parent11fd0a6fb756b9d3f387b4901aa1ce8173c03170
common : auto-download dflash- and eagle3- HF sidecars (#25811)

* common: auto-download dflash- and eagle3- HF sidecars

Mirror the existing mtp- sidecar logic to support auto-discovery and
download of DFlash (dflash-) and Eagle3 (eagle3-) speculative decoding
sidecars from Hugging Face repos.

Changes:
- Add --dflash and --eagle3 CLI flags to trigger sidecar download
- Add find_best_dflash() and find_best_eagle3() using find_best_sibling
- Exclude dflash- and eagle3- filenames from primary model selection
- Filter dflash- and eagle3- from cached model listings
- Wire download tasks that set speculative.draft.mparams as fallback

Assisted-by: pi:llama.cpp/Qwen3.6-27B
* docs : regen
common/arg.cpp
common/download.cpp
common/download.h
tools/cli/README.md
tools/completion/README.md
tools/server/README.md