 Alexander Korolev
|
7ee302e6e9
health: FLARE groups show up as a container on the controller
|
6 часов назад |
|
Alexander Korolev
|
cdd141de82
flare: mirror the package onto the controller and launch from the local copy
|
7 часов назад |
|
Alexander Korolev
|
6e4cf33a15
web: FLARE controller is discovered (not configured), dead groups are not running
|
9 часов назад |
|
Alexander Korolev
|
5eaec401fd
web: flare groups fold into the controller node only (split helpers to stay under the line cap)
|
11 часов назад |
|
Alexander Korolev
|
3f076dbe59
web: fold flare orbit groups into the controller node only
|
11 часов назад |
|
Alexander Korolev
|
04d383dc2e
web: FLARE groups show up in the dashboard like docker containers
|
11 часов назад |
|
Alexander Korolev
|
56849d5137
flare driver: parse the package path from download output, controller-only cluster preflight
|
16 часов назад |
|
Alexander Korolev
|
310fae8492
spec: flare delivery — APT is a daily snapshot (package was ~26 h old), sources stay on one build host
|
20 часов назад |
|
Alexander Korolev
|
43d40a41b5
flare driver: orbit node-name mapping, readiness wait, logs attach
|
21 часов назад |
|
Alexander Korolev
|
0a4940df8c
night log: disk audit — stopped containers pinned 211-245 G of images, root-owned leftovers, gr8/uv caches
|
21 часов назад |
|
Alexander Korolev
|
1a6126581a
spec: FLARE as a first-class engine (engine=flare, deploy=native)
|
21 часов назад |
|
Alexander Korolev
|
1544158c3d
spec: the FLARE gb10 package is our own 4-Spark contract (a test candidate), not vLLM guidance
|
21 часов назад |
|
Alexander Korolev
|
e215b55245
spec: 512K + standard alias measured (24K prompt, prefill 2,996 t/s)
|
21 часов назад |
|
Alexander Korolev
|
e450441d44
report panel: bell popup anchored, sync results as notifications, click-to-highlight, real uptime for fleet rows
|
21 часов назад |
|
Alexander Korolev
|
0d9cdda2cb
report: seed the panel from the repo cache at startup, scoped to configured repos
|
22 часов назад |
|
Alexander Korolev
|
050f0a8c5c
web: recipes cloned after startup become visible, and the panel says why a list is empty
|
22 часов назад |
|
Alexander Korolev
|
0bbd167626
night 2026-10-23: engram localization closed — 91/100, prefill parity with production
|
23 часов назад |
|
Alexander Korolev
|
cb4d7d148f
docs: AGENTS + spec — local-view measured (prefill 750 -> 3,957 t/s on the 1017 lane)
|
1 день назад |
|
Alexander Korolev
|
01011f077d
staging local-view: fix the shared-cache symlink root, nest one view per repo, adopt a pre-staged tree
|
1 день назад |
|
Alexander Korolev
|
cd5e76112a
staging: local-view kind — keep a hot side table on local NVMe, gated by a pre-start free-space check
|
1 день назад |
|
Alexander Korolev
|
72358e130c
spec: stock vLLM/SGLang vs ours for DeepSeek-V4.1-Flash-1017
|
1 день назад |
|
Alexander Korolev
|
12e79273ef
scripts: flare-upstream-watch — FLARE moves daily, check it before every experiment
|
1 день назад |
|
Alexander Korolev
|
f9606ab210
night report 2026-10-22/23: full GLM-5.3 at 90/100 with MTP, DCP impossible, prefill ceiling measured, FLARE brought up, add-on 1.3.4-1.3.9
|
1 день назад |
|
Alexander Korolev
|
60ffddaa38
build: bake SM121_VERSION through a generated source, not option_env alone
|
1 день назад |
|
Alexander Korolev
|
89553b26ac
build: rebuild when SM121_VERSION changes (option_env! alone does not invalidate cargo's cache)
|
1 день назад |
|
Alexander Korolev
|
a9cffba5fa
web: build-version badge under the dashboard title
|
1 день назад |
|
Alexander Korolev
|
b984f59269
night plan 2026-10-22/23: reproduction of the long-prefill host hang, the three envelopes to measure (KV-per-token, prefill scratch, spec depth), and the guardrails
|
1 день назад |
|
Alexander Korolev
|
4c4e608ec1
report/web: model-column copy + Hugging Face buttons, and a favicon
|
1 день назад |
|
Alexander Korolev
|
73b27de372
spec: FLARE bring-up — DeepSeek-V4.1-Flash-1017 served by NVIDIA's flare on 4 Sparks
|
2 дней назад |
|
Alexander Korolev
|
87e885d967
night report 2026-10-21/22: dashboard fixes 1.3.0-1.3.3, recipe-relative hooks, TF 1.0.12 blocked (MissingEntry) with full diagnosis, GLM-5.3 full MXFP4xFP8 blocked at 4-node worker init, GLM-5.3-Flash 2x repro 95/100
|
2 дней назад |
