 Alexander Korolev
|
2401b34ef7
ling/step5 recipes: point at the downloaded weights (Ling fp8 original 535G, Step-5 NVFP4 368G on NAS) + entrypoint=[vllm,serve] (build_tail needs it or the model lands in --entrypoint); Ling 8x TP8 stock v0.31.0
|
пре 1 недеља |
|
Alexander Korolev
|
c87238bfd6
8x lane reports: kindling8x-conc-48g (2026-10-12, 95/100 anchor) run package — md/json/log/manifest/recipe.toml
|
пре 1 недеља |
|
Alexander Korolev
|
78c48793f7
prep recipes: ling-31-flash-nvfp4 (4x, BailingMoeV3 on v0.31.0, weights pending) + step5-preview-nvfp4 (8x, dedicated step-5-preview image, VL, GB10-unverified)
|
пре 1 недеља |
|
Alexander Korolev
|
201b853c79
dgpp recipes: pin [dgpp].repo/rev (3ad63d6b) for sm121 on-demand DGPP provisioning
|
пре 1 недеља |
|
Alexander Korolev
|
28a07e0509
glm53-flash-nvfp4-fp8-dgpp-2x: DGPP-engine GLM-5.3-Flash w2 recipe (engine=dgpp, deploy=native, world_size=2, [dgpp.engine] from DGPP w2 example)
|
пре 1 недеља |
|
Alexander Korolev
|
c733c0688f
qwen38-27b-nvfp4-dgpp: first-class DGPP-engine recipe (engine=dgpp, deploy=native, [dgpp.engine] knobs → generated cluster JSON, fixed API :18080)
|
пре 1 недеља |
|
Alexander Korolev
|
541c5f6dc4
glm53-full-tp4-gr8: first-class gr8-engine recipe (engine=gr8, TP=4 switched, --tp/--peers generated, fixed API :8100)
|
пре 1 недеља |
|
Alexander Korolev
|
040e9f26b6
mimo26-flash-mopd-4x-sparkring: rebase bench (2026.10.4 image, v2.8.0) — 90/100, tg 66.2/80.7 c1/c2 (+19% vs dev-20261004)
|
пре 1 недеља |
|
Alexander Korolev
|
a09e85fb54
mimo26-flash-mopd-4x-sparkring: rebase to SparkRing 2026.10.4 image (dev-20261013, e21ba936)
|
пре 1 недеља |
|
Alexander Korolev
|
6c784467fe
glm53-kindling-2x: second perf pass (post-campaign cooldown) — d0 c1 51.4 vs 39.6 hot: run-variance baseline
|
пре 1 недеља |
|
Alexander Korolev
|
71b7f321b0
glm53-kindling-2x: perf-only re-baseline (v0.31.0 image, no label needed)
|
пре 1 недеља |
|
Alexander Korolev
|
2e45fa0bca
glm53-kindling-2x: v0.31.0-base bench (v2.8.0, 97 scen) — 95/100 quality (+2 vs 93 control), perf + depth suite 16K..256K
|
пре 1 недеља |
|
Alexander Korolev
|
8678fa1883
kindling: AGENTS checklist for changing the vLLM base (overlay rebase + version guards + post-bake smoke)
|
пре 1 недеља |
|
Alexander Korolev
|
6e0b9528ac
draft-trunc guard: v0.31.0 is a release build (bare version string, no +g<sha> scm suffix) — pin the guard to the release version, keep the fail-loud semantics
|
пре 1 недеља |
|
Alexander Korolev
|
5e04b7bfe9
glm53-kindling: rebase the 23 drifted experimental/ overlays onto the v0.31.0 base (10 files carry real drift; 21 auto-merged, 2 hand-merged: rejection_sampler accept_drafts after the gather_draft_sampled() refactor, sm90 _TRITON early-return after the 3-tuple _kv_lens_host). Unrebased overlays reintroduce ddd6fbca-era import blocks on the v0.31.0 tree -> ImportError at boot
|
пре 1 недеља |
|
Alexander Korolev
|
546e7b2025
glm53-kindling: re-pin BASE to the v0.31.0 release — Docker Hub pruned the entire old nightly-<sha> scheme, the 09-26 base tag is gone and the dir was unbuildable. v0.31.0 (cut 10-02) is the first release containing the base commit; drift audit ddd6fbca..v0.31.0: all patched files unchanged, vllm#58720 landed verbatim upstream (patch dropped, verify-base keeps asserting it), flashinfer 0.7.0->0.7.0.post1, torch/NCCL/cutlass-dsl unchanged
|
пре 1 недеља |
|
Alexander Korolev
|
b655965c5f
2x control bench on v2.8.0 (97 scen, hardmode, engine-default effort): 93/100 — the same-harness reference for gr8 (89) and DGPP (88)
|
пре 1 недеља |
|
Alexander Korolev
|
989c085a91
chat template: port the merged GLM-5.3 tool-result fix (HF PR #31) — linear sort-merge pairing replaces the O(N^2) id checks; byte-identical output validated on 3000 randomized conversations; 800 parallel calls render 5.7s -> 0.023s (the old path risked vLLM's 30s render timeout). Checksums change: next serve rebuilds both images
|
пре 1 недеља |
|
Alexander Korolev
|
0acfa6c680
new-engine quality gates on v2.8.0 (97 scen, hardmode, engine-default effort): gr8 full GLM-5.3 TP=4 = 89/100 (decode 49.9 t/s code c1, prefill 1,294 cold/50.9K cached, agg c4 64.6); DGPP Qwen3.8-Flash-Next NVFP4 w1 = 88/100 (def-effort; the low-effort run 86 on 2.7.1.dev7 is a separate datapoint)
|
пре 1 недеља |
|
Alexander Korolev
|
37dc1d74a9
gitignore: per-recipe .lock run ledgers are machine-local state
|
пре 1 недеља |
|
Alexander Korolev
|
94ac882a41
glm53-kindling-4x: 30 GiB hot KV pin (measured headroom; pool 5,083,602 tok)
|
пре 1 недеља |
|
Alexander Korolev
|
606fe37cb4
glm53-kindling-8x: 48 GiB hot KV pin (measured 24-27 GiB/node headroom at 30G)
|
пре 1 недеља |
|
Alexander Korolev
|
58ca8f5034
recipes: glm53-kindling-8x — TP8 on the basic model (b12x load-time head padding, native DFlash2 drafter)
|
пре 1 недеља |
|
Alexander Korolev
|
f1c95cde2d
README: cluster config is the add-on's sm121_config_extra option (no second repo)
|
пре 1 недеља |
|
Alexander Korolev
|
f24e55b2eb
README: no cluster config in this (public) repo — it lives in the private cluster-config repo
|
пре 1 недеља |
|
Alexander Korolev
|
df46bb8b07
docker-images: sparkring-shared
|
пре 1 недеља |
|
Alexander Korolev
|
ad8fb377bf
docker-images: solar-open2-nvfp4
|
пре 1 недеља |
|
Alexander Korolev
|
24a8341fad
docker-images: qwen38-flashnext-nvfp4
|
пре 1 недеља |
|
Alexander Korolev
|
4d470eb805
docker-images: mimo26-spark-vllm
|
пре 1 недеља |
|
Alexander Korolev
|
760cb3e782
docker-images: mimo26-spark
|
пре 1 недеља |
