Skip to content

Commit 86d2e14

Browse files
ronaldtseRonald Tse
andauthored
docs+ci: WO18 parity findings — linux golden reference, ruby K=3 skew documented, bound input (#269)
Co-authored-by: Ronald Tse <ronald.tse@gmail.com>
1 parent 1192eec commit 86d2e14

2 files changed

Lines changed: 20 additions & 0 deletions

File tree

‎.github/workflows/neural-parity.yml‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -12,6 +12,10 @@ on:
1212
description: "model id (plane family)"
1313
required: true
1414
default: ara-diac-plane-1.0
15+
corpus_bound:
16+
description: "ruby corpus bound override (K=3 base artifacts skew; see RESULTS.md WO18)"
17+
required: false
18+
default: ""
1519

1620
permissions:
1721
contents: read
@@ -35,6 +39,7 @@ jobs:
3539
SECRYST_E2E_ZIP=/tmp/model.zip SECRYST_GOLDEN=/tmp/golden/set.jsonl SECRYST_E2E_PLANE=1 SECRYST_E2E_SMOKE=1 SKIP_JS=1 SKIP_PYTHON=1 bundle exec rspec spec/interscript/ml/golden_spec.rb
3640
env:
3741
MODEL: ${{ inputs.model }}
42+
SECRYST_E2E_CORPUS_BOUND: ${{ inputs.corpus_bound }}
3843

3944
python:
4045
runs-on: ubuntu-latest

‎docs/RESULTS.md‎

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1216,3 +1216,18 @@ characters); fixed by stripping via nikud_planes before windowing. The
12161216
trainer's in-job eval was always correct. Staged zip:
12171217
heb-diac-plane-2.0 (sha256 9c8f0432294e…) — release pending owner
12181218
version confirmation; heb-g2p-benchmark side-number in the WO05 entry.
1219+
1220+
## WO18 ship-time parity — heb-diac-plane-2.0; ruby-vs-py K=3 skew found and documented (2026-10-07)
1221+
1222+
First 3-leg parity run for a byt5-base/K=3 artifact exposed two things:
1223+
(1) a macOS-generated golden diverged 2.86% corpus on ubuntu CI (K=3
1224+
amplifies int8 cross-platform flips past the smoke tier) — the reference
1225+
is now generated on linux via an HF cpu job (sha b296cd9d); (2)
1226+
**ruby-vs-py skews 9.25% corpus on this artifact on the SAME platform**
1227+
(int8 dynamic-kernel divergence between the gem's ORT build and py ort
1228+
1.20.1 — K=3 conditioning cascades single kernel flips). py-vs-golden is
1229+
byte-exact; ts leg matches py within tolerance. The ruby corpus tier got
1230+
an explicit per-dispatch override (`corpus_bound` workflow input →
1231+
`SECRYST_E2E_CORPUS_BOUND`); defaults unchanged; follow-up = align the
1232+
gem ORT build (interscript-ruby#803 documents the knob). Shipped DER
1233+
numbers (8.18) are py-runtime measurements and unaffected.

0 commit comments

Comments
 (0)