vrr/BigMoeOnEdge
2
0
Fork 0
mirror of https://github.com/Helldez/BigMoeOnEdge.git synced 2026-10-14 20:30:54 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

docs(bench): tell contributors to check the mode line before reporting a number (#190)
Some checks are pending
ci / changes (push) Waiting to run
Details
ci / format (push) Blocked by required conditions
Details
ci / host-linux (push) Blocked by required conditions
Details
ci / host-windows (push) Waiting to run
Details
ci / host-macos (push) Waiting to run
Details
ci / android-apk (push) Waiting to run
Details

Browse source
This commit is contained in:
Raffaele 2026-08-29 21:01:34 +02:00 • committed by GitHub
parent 674e7dc136
commit 4c02169fee
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 5 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

5
docs/community-benchmarks.md
View file

@ -82,6 +82,11 @@ Flags after the model path reach `bmoe-cli` verbatim (`--no-think` for gpt-oss,
default row rather than replacing it. To understand the knobs first, see
[benchmark-method.md](benchmark-method.md).
Whatever you run, check the `mode:` line before reporting a number. It says `expert streaming` when
the run measured this engine; if it says `mmap`, `--moe-stream` was missing and the run measured
stock llama.cpp instead. A row like that is still welcome as a baseline, it just has to be labelled
as one.
### On a phone
`bench-report.sh` is Linux and macOS only. On Android, push `bmoe-cli` (from