 Alexander Korolev
|
2608d645c2
AGENTS.md: document ensure-powered power on + persisted web jobs registry
|
2 tuần trước cách đây |
|
Alexander Korolev
|
d67d9d8ffa
web: jobs registry persists across restarts (web-jobs.json next to the global config; in-flight jobs marked failed with a runlock hint on boot); FTL full parity — 209 keys in all 6 locales, 19 health/exec/doctor/messages translated per locale
|
2 tuần trước cách đây |
|
Alexander Korolev
|
162b9417a5
power.rs back under the 500-line cap: HA entity toggling (toggle_one/toggle_entities/extra_ids/call) moved to power/toggle.rs, plug decisions (OnAction/decide_on/stale_nodes) to power/plug.rs; stale doc comment cleaned
|
2 tuần trước cách đây |
|
Alexander Korolev
|
3acc8fa855
power on: ensure-powered — nodes believed down with an already-on plug get a 30 s plug power cycle (2-5 s cuts left the PSU rail energized and the machine in soft-off); per-node outcome report becomes the job message; CLI uses a 1.5 s TCP reachability probe, web uses the last snapshot
|
2 tuần trước cách đây |
|
Alexander Korolev
|
9cf3deb380
web logs: ANSI SGR rendering in log windows — themed palette (8 colors/bold/dim/italic/underline, fg replaces fg), strip OSC + non-SGR CSI, carriage-return progress frames collapse to the last frame, debounced batched re-render, raw text kept for copy; node probe + playwright timing fix
|
2 tuần trước cách đây |
|
Alexander Korolev
|
b9da59a70f
web token: viewer links in the embedded recipes panel carry the auth token (query fallback + localStorage) — bare /r/ links 401'd in token mode
|
2 tuần trước cách đây |
|
Alexander Korolev
|
b5cf4a4ba8
web topo: cap frame pads (5-level nesting pushed the outer frame above the canvas — the 'top frame cut off'), draw the map at container width with scroll, add a persisted cluster filter (hidden clusters recompute the nest depth)
|
2 tuần trước cách đây |
|
Alexander Korolev
|
5eecadc573
web: stop works off the recipes actually running on the cluster — single running recipe stops directly (no picking), several get a picker of running ones only, container list names the node; serve-form fallback kept for up-but-empty clusters
|
2 tuần trước cách đây |
|
Alexander Korolev
|
6ccb448146
web: fix two-step confirms (power on/off) doing nothing — openDialogHtml stored the placeholder go instead of opts.go, so Confirm closed the dialog without firing the request; playwright regression covers the full power-on flow
|
2 tuần trước cách đây |
|
Alexander Korolev
|
1421e95250
campaign 2026-10-03: MiMo 2x/Pro 8x benched (pluto/wired), kindling 2x+TP6, karmic-kraken image, depth-hang forensics
|
2 tuần trước cách đây |
|
Alexander Korolev
|
fef19db3bb
bench: MiMo 4x kraken full re-run via pluto (wired) — depth suite now works
|
2 tuần trước cách đây |
|
Alexander Korolev
|
4a99ace3ce
fix(bench): --ssh-host depth suite — create the remote scratch dir before cd
|
2 tuần trước cách đây |
|
Alexander Korolev
|
cad1944a7d
bench: kindling 4x + MiMo 4x kraken via pluto (wired) — WiFi-isolation campaign
|
3 tuần trước cách đây |
|
Alexander Korolev
|
b70649a547
feat(bench): --ssh-host — run tool-eval-bench on a remote host over SSH
|
3 tuần trước cách đây |
|
Alexander Korolev
|
034c9b848f
recipes/mimo-*-sparkring: bump sparkring image to dev-20261001-kraken
|
3 tuần trước cách đây |
|
Alexander Korolev
|
7559d62476
recipes/mimo-*-sparkring: float the model id + dflash drafter
|
3 tuần trước cách đây |
|
Alexander Korolev
|
c2871f857a
kindling 4x re-bench on 27cf839 image (label kindling-4x-27cf839): 94/100 (166/176), identical to the a3c3a1d baseline — same fails TC-43/51/68/80, same depth curve. The megamoe #67 + spec-config #64 in the new image neither regress nor improve GLM-5.3-Flash 4x quality. Model now floats via refs/main (da920bb0 at bench time); manifest predates the models-field, so it carries no snapshot (next bench will).
|
3 tuần trước cách đây |
|
Alexander Korolev
|
24d6737bc3
de-pin model revisions: kindling 2x/4x model -> floating org/model id (engine resolves via NAS refs/main -> latest snapshot, same as every other recipe; the pinned da920bb0 stays as an informational comment); 6x notes its tp6pad source rev. Bench manifest now records the model's resolved refs/main snapshot so a floating model stays reproducible (backward-compatible #[serde(default)]).
|
3 tuần trước cách đây |
|
Alexander Korolev
|
729f454eaa
kindling 4x: full campaign report package (md/json/log/manifest/recipe snapshot) — 94/100 quality, depth curve to 256K with no decode cliff (tg 60-88 t/s), TTFT 2.1s @256K c1
|
3 tuần trước cách đây |
|
Alexander Korolev
|
f944aea35c
kindling 4x: first full campaign — quality 166/176 (94/100, best on cluster), perf table, boot bugs ledger (spec-config form, UMA gate coin flip, control-file dirs); spec section + report artifacts (json+log snapshot, depth suite in flight)
|
3 tuần trước cách đây |
|
Alexander Korolev
|
b58b87fa8f
skill serve-boot-triage: 4b — long-running local work (benches) dies under the tool harness; move it to a cluster node (build sm121 there, tmux session, host-side tokenizer path, report copy-back)
|
3 tuần trước cách đây |
|
Alexander Korolev
|
bfbbf854b3
bench: the script-pty wrapper is macOS-only — util-linux script (Linux nodes) has no positional command and parsed the harness flags as its own ('unrecognized option --base-url'); non-macOS goes to the direct-exec fallback
|
3 tuần trước cách đây |
|
Alexander Korolev
|
e6204c6b95
kindling recipes: control paths are FILES, not the cache dirs — VLLM_{ADAPTIVE_K,DRAFT_TRUNC}_CONTROL pointed at the mounted dir roots, and their _read_control() does os.stat+open on the path (stat succeeds on a dir, open does not) => IsADirectoryError killed the engine core right after KV allocation; point them at <dir>/control (absent file = safe defaults: tau 0.3 / adapt, written file persists + live-control)
|
3 tuần trước cách đây |
|
Alexander Korolev
|
571a5499f1
history: 2026-10-02 night — kindling 4x e2e (spec-config doubling, triple-quote form, upstream 27cf839 re-vendor, GB10 UMA gate coin flip fixed via GMU 0.86 + last-moment drop_caches mod), serve-boot-triage skill, headless dgx6/7/8, upstream sweep (MiMo/GLM/DFlash2 pins current, eugr nightly-20261001 new)
|
3 tuần trước cách đây |
|
Alexander Korolev
|
2190f53141
kindling recipes: kill the GB10 UMA startup gate coin flip — GMU 0.88->0.86 (required 104.60 vs observed free 106.28-106.92 at worker init; the container stack refills ~11 GiB between the pre-container drop_caches and the check) + last-moment drop_caches mod right before the engine exec; the hard --kv-cache-memory pin is untouched so the KV pool is unchanged
|
3 tuần trước cách đây |
|
Alexander Korolev
|
6cb6d346ed
skill: serve-boot-triage (log-preserving --keep launches, detached serve, per-node state matrix, hard rules incl. config IPs, GB10 failure patterns); revert the redundant recipe drop_caches hooks (launcher config default is already true)
|
3 tuần trước cách đây |
|
Alexander Korolev
|
b23cc1ec1f
kindling recipes: drop_caches before_serve — the 0.88 GMU startup gate reads MemFree on GB10 UMA and random nodes carried ~15 GiB of weight/NFS page cache (106.53 free vs 107.03 required on rank1, the '2 of 4 nodes' boot failure); reclaim cache per node before containers start
|
3 tuần trước cách đây |
|
Alexander Korolev
|
39c8ae1432
kindling: re-vendor to upstream 27cf839c2133 (spec-config moved to entrypoint/SPEC_EXTRA #64 — same bug we hit; megamoe routed-expert-0 fix #67; kindling.json results #68); README notes B-mode equivalence
|
3 tuần trước cách đây |
|
Alexander Korolev
|
48af96bc5d
kindling recipes: close the --speculative-config triple-quote on the content line (mimo shape) — a trailing newline before the closing hit a different code path and the engine received apostrophe-wrapped JSON (argparse 'cannot be converted to loads'); form now byte-identical to the working mimo recipes, argv validated in-image against the real vLLM
|
3 tuần trước cách đây |
|
Alexander Korolev
|
4c3f95d199
kindling recipes: fix doubled --speculative-config value (generator inserted the JSON twice; engine boot crashed with argparse 'cannot be converted' — caught by first e2e diag run)
|
3 tuần trước cách đây |
