 Alexander Korolev
|
ba2d898771
docs(AGENTS): mikrotik power-off relay semantics
|
3 週間 前 |
|
Alexander Korolev
|
5c82201e40
power: mikrotik off must cut the outlet relay too
|
3 週間 前 |
|
Alexander Korolev
|
c01f43cdc8
bench(engram-local): aiden16 localized-Engram verification — 91/100, pp 10x, penalty gone
|
3 週間 前 |
|
Alexander Korolev
|
b07f853b50
docs: aiden16 localized-engram cutover + liveness fix shift note
|
3 週間 前 |
|
Alexander Korolev
|
6ae3c5159c
readiness: watch() must not call an API that never opened 'dead'
|
3 週間 前 |
|
Alexander Korolev
|
789983fc62
staging(sparse-copy): aiden base+rem row tiling + aiden16 localized engram staging
|
3 週間 前 |
|
Alexander Korolev
|
bc269457c2
bench(exl3-staged): 92/100 declarative-staging verification + bench --tokenizer passthrough
|
3 週間 前 |
|
Alexander Korolev
|
e586bd2416
staging: EXL3 live-cutover fixes — src from config, syntactic link resolution, ro-safe export placeholders
|
3 週間 前 |
|
Alexander Korolev
|
f9582dd215
staging(hub-mirror): resolve source links syntactically, not realpath
|
3 週間 前 |
|
Alexander Korolev
|
05d095ac96
staging: sparse-copy format (aiden16 DSV41_ENGRAM_DIR contract) + packed-Mia bench notes
|
3 週間 前 |
|
Alexander Korolev
|
952a841abd
EXL3 migrated to declarative staging (+ export materialization)
|
3 週間 前 |
|
Alexander Korolev
|
feb7a0fa88
night-shift note: packed-Engram cutover — decode c1 29 → 72-74 t/s
|
3 週間 前 |
|
Alexander Korolev
|
f531496105
staging P2: sparse-rows packed Engram — per-rank local shards, ledger-verified
|
3 週間 前 |
|
Alexander Korolev
|
58c6892537
staging P1: declarative [[staging]] hub-mirror, ledger-verified
|
3 週間 前 |
|
Alexander Korolev
|
766b7fb748
mia-4x lane live: production line armed end-to-end (boot-log verified), API up, c1 decode 29 t/s on unpacked NAS engram — staging P2 is the next lever
|
3 週間 前 |
|
Alexander Korolev
|
4f9cf636eb
mia-4x: env TP_SIZE/EP_SIZE/DSV41_TP_PAD — adapters read world shape from env, tp3_pad misfired at TP4 (24-head sparse-MLA abort)
|
3 週間 前 |
|
Alexander Korolev
|
6a763946d0
staging spec: image-GC LRU (user: docker images must evict too) + Mia TP4 lane vendored (docker-images/mia-tp4 @ cad252b) + recipe dsv41-flash-4x-mia
|
3 週間 前 |
|
Alexander Korolev
|
3da42bc210
staging-feature spec (P1 hub-mirror, P2 sparse-rows, P3 eviction, P4 probe-and-go) + 09-27 shift: exl3 local bench 92/100, Mia TP4 discovery
|
3 週間 前 |
|
Alexander Korolev
|
f098a21e59
exl3-2x: node-local weight staging done + page15 NAS-fallback binds + per-run receipt clean; decode 13.3 -> 35 t/s
|
3 週間 前 |
|
Alexander Korolev
|
c31f0f107f
shift notes: qwen pinned-image verdict — bench 94, live agent clean, dgx3 off
|
3 週間 前 |
|
Alexander Korolev
|
280f4d0c1f
bench: qwen3.8 int4 1x re-run on pinned b12x (nightly-20260913) — 94/100 (164/174), no regression vs the suspect-image run
|
3 週間 前 |
|
Alexander Korolev
|
4d7b927931
shift notes: qwen garble root cause — unpinned b12x image (nightly-20260925), forum 382733 #195-198; pinned to nightly-20260913
|
3 週間 前 |
|
Alexander Korolev
|
35cb002a16
qwen3.8: restore bench baseline params (MTP k=4, marlin atomic-add 1) on the pinned image — both were garble red herrings
|
3 週間 前 |
|
Alexander Korolev
|
3d067b1081
fix: image presence check accepts digest refs (docker image ls does not); pin qwen3.8 b12x to the validated nightly-20260913 digest (floating latest 20260925 garbles long contexts, forum 382733 #195-198); min-recipe A/B lane (no MTP, no prefix cache)
|
3 週間 前 |
|
Alexander Korolev
|
22578fa25f
bench: DSV4.1-Flash EXL3 3bpw 2x — 164/176 (93), tg 13.3 t/s @c1; safety pass (injection ignored)
|
3 週間 前 |
|
Alexander Korolev
|
5d3493a615
shift notes: EXL3 O_NOFOLLOW root cause + manifest-mount fix
|
3 週間 前 |
|
Alexander Korolev
|
bf593fd4fd
test: add extra_mounts_manifest: None to Recipe test literals (fix test-profile build)
|
3 週間 前 |
|
Alexander Korolev
|
08f53fc875
exl3: per-file bind mounts via extra_mounts_manifest — the runtime opens shards with O_NOFOLLOW, so a symlink final component => ELOOP (deterministic, nothing to do with NFS); manifest generated per node by before-serve hook, launcher reads it over ssh after the hooks. EXL3 2x now boots and serves
|
3 週間 前 |
|
Alexander Korolev
|
a1472225ad
refactor: move prebuilt-image check to sync::ensure_prebuilt_image (mod.rs back under 500)
|
3 週間 前 |
|
Alexander Korolev
|
89c21cc236
fix: stop --cluster narrows multi-deployment (no more 'use --node' when --cluster given); --node without --cluster resolves via node containment. prebuilt images (no docker_image_dir) are presence-checked + node-to-node synced before docker run — no more silent Docker Hub pull + false 'started'
|
3 週間 前 |
