Alexander Korolev
|
8f891c6358
docs: dgx4 NVMe investigation - DRAM-less FTL thrash, not a fault
|
11 годин тому |
Alexander Korolev
|
2977785ed3
docs: current FLARE lane state, uniform-binary rule, refreshed upstream pin
|
11 годин тому |
Alexander Korolev
|
06b53e0c78
flare: the dashboard/probe controller pin also resolves outside the lane cluster
|
12 годин тому |
Alexander Korolev
|
7271c95ae2
release: sm121 0.2.7 — FLARE runtime override, controller outside the launch cluster
|
12 годин тому |
Alexander Korolev
|
0ebe2ce33e
flare: prod-no4 is production - dgx4 fails sustained Engram duty even with the reader patch
|
12 годин тому |
Alexander Korolev
|
7d0dc801cd
flare: resolve the pinned controller across clusters; prod-no4 lane config
|
14 годин тому |
Alexander Korolev
|
cc0c08393f
flare: first-class --executable override + Engram-stall post-mortem
|
23 годин тому |
Alexander Korolev
|
a1d58a7374
docs(AGENTS): FLARE engine deploy, orbit/controller mapping and the two-store disk budget
|
2 днів тому |
Alexander Korolev
|
f3548d85e8
flare: honour the pinned orbit controller and match orbit node names by name (0.2.6)
|
2 днів тому |
Alexander Korolev
|
017fa761f2
flare: mirror blob-store packages with hardlinks; adopt an already-downloaded package (0.2.6)
|
2 днів тому |
Alexander Korolev
|
fd3bcb0ef5
docs: Mistral Large 4 8x serving notes + bringup spec (vLLM vs SGLang A/B)
|
5 днів тому |
Alexander Korolev
|
d67c8e2e10
release: sm121 0.2.5 — sync waits for the containerd snapshot unpack (exec probe per peer)
|
5 днів тому |
Alexander Korolev
|
be1da39520
release: sm121 0.2.4 — cached liveness, TCP switch ping, bench log/job fixes
|
6 днів тому |
Alexander Korolev
|
dd8dc57843
release: sm121 0.2.3 — FLARE all-rank topology and orbit attach logs
|
1 тиждень тому |
Alexander Korolev
|
ecb1eec8ff
docs: dashboard bench window (params, node, live log, cancel, reopen)
|
1 тиждень тому |
Alexander Korolev
|
d8dc94acad
release: sm121 0.2.2 — dashboard bench window (params, node, live log, cancel, reopen)
|
1 тиждень тому |
Alexander Korolev
|
79848764a2
docs: FLARE prod uses no --roce-ring; legacy CX7 ports were the cause
|
1 тиждень тому |
Alexander Korolev
|
1bf1c11ce4
release: sm121 0.2.1 — FLARE pairing, package skip/rsync, settings check
|
1 тиждень тому |
Alexander Korolev
|
5ab0becb8d
dsv41 flare: record recipe-run bench and alias state
|
1 тиждень тому |
Alexander Korolev
|
6c3a20d976
flare: long-running package/luanch ssh; ledger by recipe dir
|
1 тиждень тому |
Alexander Korolev
|
b134c7f539
release: sm121 0.2.0; changelog and release process
|
1 тиждень тому |
Alexander Korolev
|
209bdbebad
dsv41 flare: record v3.0.0 bench of 2026-10-27 package
|
1 тиждень тому |
Alexander Korolev
|
77d42f79e6
dsv41 flare: document 2026-10-27 APT package, c32, NAS staging
|
1 тиждень тому |
Alexander Korolev
|
51f45ea9d2
flare: handle cached packages and nested package companions
|
1 тиждень тому |
Alexander Korolev
|
ce3ab352bf
bench: pin tool-eval-bench v3.0.0 as the new standard; v2.8.0 reports are historical
|
1 тиждень тому |
Alexander Korolev
|
d799a83958
specs: upstream watch 2026-10-26 — FLARE head, tool-eval-bench v3.0.0, GR8/DGPP
|
1 тиждень тому |
Alexander Korolev
|
fc3b948563
specs: prod moved to FLARE — fabric ring, extra RoCE rails, controller identity
|
1 тиждень тому |
Alexander Korolev
|
5b019795f9
flare on the prod cluster: fabric ring, ledger, per-cluster probe
|
1 тиждень тому |
Alexander Korolev
|
fc460279dc
web: cache + settings windows; cache update is one code path
|
1 тиждень тому |
Alexander Korolev
|
013b074b62
specs: resident-scales A/B measured (+41% prefill, no d8192 decode sag, ~1.5 GB/rank)
|
1 тиждень тому |