frank-primary: keep 1.0.13 engine pin — 1.2.0 bench rejected (accept 0.7123 vs 0.7549, decode -2.9%; vram pair exonerated, ledger-exact)

This commit is contained in:
Frank
2026-10-06 21:35:39 -07:00
parent 8c887e6fc9
commit 910c54b6ff
+17 -9
View File
@@ -51,15 +51,23 @@
# --num-speculative-tokens 3.
# - The fused TP all-reduce is declared over max-num-seqs x
# (1 + --num-speculative-tokens) rows; 3 x 3 = 9 rows fits either wire.
# Engine pin: Codeberg tag v1.2.0 (commit e95d397), Docker Hub tag 1.2.0
# (no `v` prefix on tags), index digest below, pushed 2026-10-06. 1.2.0
# REMOVED --expert-vs-cache-ratio in favour of the explicit
# --vram-weights-mib / --vram-kv-mib pair (core/config.cpp), and added
# MXFP4 dense kernels, --p2p host twins, a stop-string fix, a DFlash2
# fix, and a MoE weight-rotation change (92b8aaf). Prior pin
# sha256:4c3f484564bb2872784e46b0a59cbd71becc9c1e746e301a3596ffae253c322e
# = tag 1.0.13, the rollback engine.
FRANK_PRIMARY_IMAGE=stilldeadcode/radiance@sha256:665775ed1d90de6dbb25558e07fc60d20328f1346f96f3c6e00845eb763f01fa
# Engine pin: Codeberg tag v1.0.13 (commit 53282fa), Docker Hub tag 1.0.13,
# index digest below. REJECTED UPGRADE: 1.2.0 (sha256:665775ed1d90de6dbb
# 25558e07fc60d20328f1346f96f3c6e00845eb763f01fa, tag 1.2.0, commit
# e95d397) was benched 2026-10-06 on this exact shape (fp8-native rad,
# vram pair, spec 2, seqs 3, 43.6k greedy probe): decode 29.21 t/s and
# draft acceptance 0.7123 vs the 1.0.13 control's 30.08 and 0.7549.
# Acceptance on the greedy harness is deterministic (the control
# reproduced the recorded 0.755 exactly), so the drop is a real numerics
# change — the 1.2.0 MoE weight-rotation change (92b8aaf) is the named
# suspect; decode loss tracks the acceptance ratio (0.943 vs 0.945).
# The vram-pair translation itself is exact (1.2.0 boot ledger took
# vram_weights 25.46 GiB / vram_kv 3.46 GiB per card byte-identical to
# the 1.0.13 ledger at EVC 0.82). Re-test 1.2.0+ when the rotation change
# is untangled; 1.2.0 also REMOVED --expert-vs-cache-ratio in favour of
# the --vram-weights-mib / --vram-kv-mib pair (core/config.cpp) — this
# file already uses the pair, which BOTH pins accept (rollback symmetry).
FRANK_PRIMARY_IMAGE=stilldeadcode/radiance@sha256:4c3f484564bb2872784e46b0a59cbd71becc9c1e746e301a3596ffae253c322e
# The .rad container FILE (mounted under /models) and the state dir for the
# prefix-cache disk tier (mounted rw at /data). The model dir holds ONLY
# .rad files; the raw HF download and its staging dir live beside it and