BigMoeOnEdge/scripts
Helldez d9381f6363 docs: benchmark pages as a contributor guide, and the Apple platform limits
The benchmark call is pinned, and both pages were written for the maintainer
rather than for the people landing on them. community-benchmarks.md opened with
a list of hardware we want, which reads as an entry requirement, and neither
page ever answered the first question a contributor has: what do I set?

community-benchmarks.md:
- a "start here" for the three cases someone is actually in (PC or laptop,
  Android phone, Apple hardware), each with the command and what to paste back
- the settings-override table, which used to be one buried sentence
- an explicit adb protocol for phones
- "hardware we want to see" moved to the end as open questions: any hardware is
  a useful row, the list is what we cannot answer ourselves

benchmark-method.md:
- reference device out of the opening, named once at the end as the provenance
  of the published numbers
- new "choosing the parameters": lossless, lossy and experimental knobs kept in
  separate tables, each with its default and the telemetry field that says
  whether moving it worked
- the hard-won rules kept as method rather than as the story of one session

The community protocol now pins --ubatch 512, which the app has always done and
bench-report.sh never did: prefill width costs resident memory the expert cache
would otherwise get, so a host row was running a different configuration from
the app it is compared against. UBATCH= overrides it.

Two platform limits documented for the first time: macOS has no O_DIRECT and
the engine does not call the F_NOCACHE equivalent, so expert reads there go
through the page cache while the metrics still report o_direct=1; and there is
no iOS target at all. Both were already true.
2026-08-28 16:53:22 +02:00
..
bench-analyze.py chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-lib.ps1 chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-matrix-rework.ps1 chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-matrix.ps1 chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-pr23-c2000.ps1 chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-pr23-summary.py chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-prefetch.ps1 chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00
bench-report.sh docs: benchmark pages as a contributor guide, and the Apple platform limits 2026-08-28 16:53:22 +02:00
bench-run.sh Mark the two doc-invoked host scripts executable (#164) 2026-08-25 11:41:24 +02:00
bench-warmonly.sh Mark the two doc-invoked host scripts executable (#164) 2026-08-25 11:41:24 +02:00
build-android.ps1 build(android): stage an explicit library list, and force GGML_OPENCL off (#124) 2026-07-28 15:10:49 +02:00
build-host.sh Mark the two doc-invoked host scripts executable (#164) 2026-08-25 11:41:24 +02:00
decode-analyze.py feat(trace): add layer-granularity compute trace (--compute-trace-layers) 2026-07-19 09:21:39 +02:00
gptoss-matrix.sh Mark the two doc-invoked host scripts executable (#164) 2026-08-25 11:41:24 +02:00
gptoss-mmap.sh Mark the two doc-invoked host scripts executable (#164) 2026-08-25 11:41:24 +02:00
inspect-hf-moe-release.py feat(moe): Qwen3.8-Flash-Next support (#172) 2026-08-28 10:07:43 +02:00
make-tiny-moe.py feat(moe): stream split multi-shard ggufs natively + DeepSeek V4 Flash recipe (#144) 2026-08-01 23:34:35 +02:00
route-analyze.py feat(moe): --drop-cold-experts — spend quality only where it buys I/O (#95) 2026-07-22 17:21:55 +02:00
route-drop-replay.py feat(moe): --drop-cold-experts — spend quality only where it buys I/O (#95) 2026-07-22 17:21:55 +02:00
route-replay.py fix(tools): make the diagnostics fail instead of reporting a confident wrong number 2026-07-20 12:25:58 +02:00
route-viewer.py feat(scripts): route-viewer.py — read a route trace without a spreadsheet 2026-07-15 09:52:09 +02:00
trace_io.py chore(scripts): consolidate the bench drivers, prune the retired framing 2026-07-17 10:41:13 +02:00