 Alexander Korolev
|
cbf8c38243
sparkring lane: recipe transcribed from their published per-rank compose (installer image digest, toolchain serve, full env) with switch-only deviations noted
|
3 недель назад |
|
Alexander Korolev
|
6b16590b1f
sync: one docker save + parallel fabric fan-out (3x on a 4-node set), inner hop key = node-side ~/.ssh/<basename> (mac path broke dgx5-8), known_hosts lives until task end (10G fallback resurrected); sm121 sync-image command
|
3 недель назад |
|
Alexander Korolev
|
0ee032e4e6
bench: depth-suite phase (16K/64K/128K/256K @ c2, 2 repeats) appended to the campaign log; report/tests.rs split (perf_box_tests) for the line cap
|
3 недель назад |
|
Alexander Korolev
|
6b75b3427b
morning shift: stop/serve race fixed + JIT cache persistence notes; lane left live for prod
|
3 недель назад |
|
Alexander Korolev
|
b6aed54ef8
stop: verify absence in ANY state (running+exited) then rm, second running-only loop; serve: up to 3 re-probes on short-free verdict
|
3 недель назад |
|
Alexander Korolev
|
a5b75d3014
stop: explicitly rm labeled stopped containers + 1s poll until no running labeled containers — closes the post-stop 'occupied' race that failed back-to-back serves
|
3 недель назад |
|
Alexander Korolev
|
a030038c21
mimo-vllm: persist inductor+triton caches (TORCHINDUCTOR_CACHE_DIR/TRITON_CACHE_DIR + triton mount) — kills TritonBundler cubin-not-found churn
|
3 недель назад |
|
Alexander Korolev
|
e3f7dbc229
mimo26-vllm: sparkring-contract bench 170/184 (92/100) + c2 long-context collapse logged for morning
|
3 недель назад |
|
Alexander Korolev
|
06590549ff
mimo-vllm: drop sparkring 20GiB KV pin — GB10 UMA auto-sizing keeps ~54GiB pool; pin would strand 34GiB
|
3 недель назад |
|
Alexander Korolev
|
1f59f0586e
mimo-vllm: adopt sparkring tp4 serve contract (spec5, 16x8192, 20GiB KV pin, bounce loader, BLHNC, A16 lm-head, extended-GID prefill); ring-only transport NOT ported (we're switched)
|
3 недель назад |
|
Alexander Korolev
|
9b9109112f
mimo: vLLM lane serving — 168/184 parity + slope sweep flattens the 256K cliff (0.30→0.74 of d0) at −34% d0; shift log 09-28
|
3 недель назад |
|
Alexander Korolev
|
1440b6e841
mimo26-vllm: quality bench vllm-kk-nightly-dflash — 168/184 (91/100), parity with sglang anchor
|
3 недель назад |
|
Alexander Korolev
|
4a33f594df
mimo-vllm: garble root cause = instanttensor mis-maps MiMo split-GDN weights; b12x loader contract (mm-encoder-tp-mode weights, full Omni, bf16 KV, VLLM_PLUGINS=b12x_loader); dflash re-enabled per Pro launcher; quality bench started
|
3 недель назад |
|
Alexander Korolev
|
787ea12644
mimo-vllm: spec RESOLVED note (nightly fixes all four walls); lane live, quality bench running
|
3 недель назад |
|
Alexander Korolev
|
c4668aa86f
mimo-vllm: reopen lane on eugr/spark-vllm-b12x nightly (karmic-kraken: MiMo aux, BLHNC layouts, b12x noncausal, eagle3 unwrap) + b12x linear/moe, CUTE_DSL_ARCH=sm_121a
|
3 недель назад |
|
Alexander Korolev
|
02846405b1
mimo: vLLM+DFlash experiment PARKED (r39 KV-layout wall); findings spec + recipe banner, patches kept for reopen
|
3 недель назад |
|
Alexander Korolev
|
1d0bb43ece
mimo26: vLLM B12X A/B lane recipe + sglang 4x fp8-KV trial rollback
|
3 недель назад |
|
Alexander Korolev
|
4ace7f4d06
bench roce-quality: 168/184 (91/100), resp 79, decode +10-15% — RoCEnante is quality-neutral
|
3 недель назад |
|
Alexander Korolev
|
615e526ecd
mimo26-flash-mopd-4x: RoCEnante live on TP4 (a15), tg +20-36%
|
3 недель назад |
|
Alexander Korolev
|
7a55726869
mimo26-spark: a15 — relax sglang RoCEnante gate from world==8 to b12x's 2..16
|
3 недель назад |
|
Alexander Korolev
|
7eba2f933f
report: bench completion is self-describing — no pool-size hardcode
|
3 недель назад |
|
Alexander Korolev
|
050f870355
mimo: 4x lane in production (GMU 0.94, bench 91/100, resp 76), JIT cache mounts
|
3 недель назад |
|
Alexander Korolev
|
3f71f5b346
mimo26-flash-mopd-4x: new 4-node contingency lane, GMU 0.94 prod (pool 4.14M)
|
3 недель назад |
|
Alexander Korolev
|
f39449c4ca
bench: --parallel passthrough + config C bench 166/184 (90/100, two-stream)
|
3 недель назад |
|
Alexander Korolev
|
ae420833fc
history: Pro-MOPD download was complete — GB/GiB misread (535 GiB == 573.5 GB), snapshot adea8e2c verified 130/130
|
3 недель назад |
|
Alexander Korolev
|
7ec09146a3
mimo26-flash-mopd-2x: config C verified — 512K context live (pool 570,752)
|
3 недель назад |
|
Alexander Korolev
|
848f902f3e
orchestrate: re-probe once before failing 'not enough free nodes'
|
3 недель назад |
|
Alexander Korolev
|
2f6c39e270
bench(mopd-kv088): MOPD config B — 168/184, pool 499K; shift notes
|
3 недель назад |
|
Alexander Korolev
|
c8676d703b
recipes(mimo-mopd-2x): mem-fraction 0.85→0.88 — KV pool 306,752→498,752 tokens
|
3 недель назад |
|
Alexander Korolev
|
6bc35921d8
bench(mopd-base): MiMo-V2.6-Flash-MOPD 2x — 170/184 (92/100, ★★★★★), +9 over Flash-RL
|
3 недель назад |
