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:
1 parent
8c887e6fc9
commit
910c54b6ff
1 file changed
+17
-9
@@ -51,15 +51,23 @@
|
|||||||
# --num-speculative-tokens 3.
|
# --num-speculative-tokens 3.
|
||||||
# - The fused TP all-reduce is declared over max-num-seqs x
|
# - 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.
|
# (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
|
# Engine pin: Codeberg tag v1.0.13 (commit 53282fa), Docker Hub tag 1.0.13,
|
||||||
# (no `v` prefix on tags), index digest below, pushed 2026-10-06. 1.2.0
|
# index digest below. REJECTED UPGRADE: 1.2.0 (sha256:665775ed1d90de6dbb
|
||||||
# REMOVED --expert-vs-cache-ratio in favour of the explicit
|
# 25558e07fc60d20328f1346f96f3c6e00845eb763f01fa, tag 1.2.0, commit
|
||||||
# --vram-weights-mib / --vram-kv-mib pair (core/config.cpp), and added
|
# e95d397) was benched 2026-10-06 on this exact shape (fp8-native rad,
|
||||||
# MXFP4 dense kernels, --p2p host twins, a stop-string fix, a DFlash2
|
# vram pair, spec 2, seqs 3, 43.6k greedy probe): decode 29.21 t/s and
|
||||||
# fix, and a MoE weight-rotation change (92b8aaf). Prior pin
|
# draft acceptance 0.7123 vs the 1.0.13 control's 30.08 and 0.7549.
|
||||||
# sha256:4c3f484564bb2872784e46b0a59cbd71becc9c1e746e301a3596ffae253c322e
|
# Acceptance on the greedy harness is deterministic (the control
|
||||||
# = tag 1.0.13, the rollback engine.
|
# reproduced the recorded 0.755 exactly), so the drop is a real numerics
|
||||||
FRANK_PRIMARY_IMAGE=stilldeadcode/radiance@sha256:665775ed1d90de6dbb25558e07fc60d20328f1346f96f3c6e00845eb763f01fa
|
# 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
|
# 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
|
# 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
|
# .rad files; the raw HF download and its staging dir live beside it and
|
||||||
|
|||||||
Reference in new issue
Block a user