Skip to content

docs(bench): drop redundant GEIST_WEIGHT_MMAP=1; fix stale spec-head default - #62

Merged
geisten merged 1 commit into
mainfrom
docs/prune-redundant-bench-flags
Jul 5, 2026
Merged

docs(bench): drop redundant GEIST_WEIGHT_MMAP=1; fix stale spec-head default#62
geisten merged 1 commit into
mainfrom
docs/prune-redundant-bench-flags

Conversation

@geisten

@geisten geisten commented Jul 5, 2026

Copy link
Copy Markdown
Owner

Small benchmark-doc cleanup — no code changes.

What

  • Drop GEIST_WEIGHT_MMAP=1 from three benchmark command lines (BENCHMARK_PI5.md, compare_ternary_pi5.sh ×2). It is already the runtime default (arch_state.c), so setting it was cargo-cult. The explanatory note that mmap-alias is the default (and why it matters on a 4 GB Pi) stays.
  • Fix a stale claim in TERNARY_BITNET.md: the i8-sketch speculative output head is now default-on for greedy decode on an eligible tied lm_head (spec_head.c), not "opt-in; default off".

Deliberately left alone

  • Prose that documents GEIST_WEIGHT_MMAP=1 as the default.
  • The predecode anti-pattern example (shows what not to set on the Pi, with measurements).
  • compare_metal.sh's GEIST_M_MAX=${…:-128} — a real tuning knob with a fallback, not redundant.

…ale spec-head default

GEIST_WEIGHT_MMAP=1 is already the runtime default (arch_state.c), so
setting it on the bench command lines was cargo-cult. Also correct
TERNARY_BITNET.md: the i8-sketch spec head is now default-on for greedy
decode on an eligible tied lm_head (spec_head.c), not opt-in.
@geisten
geisten merged commit e9d9dd7 into main Jul 5, 2026
6 checks passed
@geisten
geisten deleted the docs/prune-redundant-bench-flags branch July 5, 2026 08:21
geisten added a commit that referenced this pull request Jul 5, 2026
…72)

The metal-beat-llamacpp-plan.md close (987/81.2, 2026-07-04) predates the
#58/#60 Metal work and the #62/#63 dead-flag cleanup. Add a dated data point:
a fresh cool compare_metal.sh on main (1020d45) reads 1006 pp512 / 81.6 tg64 /
445 total vs llama same-run 1540/92.8 — within noise of the close. Confirms the
flag/kernel/replay removals did not regress Metal. No historical numbers changed.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant