Эхлэл Бүгдийг харах Тусламж
Нэвтрэх
sm121
/
recipes
2
0
Салаа 0
Файлууд Асуудлууд 0 Хуулах хүсэлтүүд 0 Мэдлэгийн сан

Тайлбар байхгүй

Мод: c5a8444bb5
Салаанууд Тагууд
recipes
Alexander Korolev c5a8444bb5 glm53-flash-fp4-local-2x: add distributed_backend="mp" + pipeline_parallel_size=1 + shm_size=16g — the missing distributed_backend made the launcher skip ALL dist flags, so each node booted an independent single-node engine loading the FULL 166 GB checkpoint (the root cause of the four stalls); launcher now refuses this config (4029b28) 6 өдөр өмнө
docker-images a972f8b214 nemotron-35-bolt-74b-24gb: switch to the CARD's image path — docker-images/nemotron-bolt-b12x (thin FROM the card's stock vLLM nightly 77154ca by arm64 digest + b12x 1.5.1 + ultra_v3 parser plugin baked at /opt); recipe flags verbatim from the card incl. LiLiCorr dflash spec-decode (7) + ultra_v3 parser; only deviation GMU 0.92→0.90 (clean Spark reports 111.55/121.63 GiB free; 0.92 misses the boot gate by 0.35). eugr/spark-vllm-b12x (03a95654) does NOT carry LiLiCorrDraftModel/ultra_v3 — it cannot serve Bolt with the card's flags. Verified live: LiLiCorr speculator captured, smoke OK. 6 өдөр өмнө
recipes c5a8444bb5 glm53-flash-fp4-local-2x: add distributed_backend="mp" + pipeline_parallel_size=1 + shm_size=16g — the missing distributed_backend made the launcher skip ALL dist flags, so each node booted an independent single-node engine loading the FULL 166 GB checkpoint (the root cause of the four stalls); launcher now refuses this config (4029b28) 6 өдөр өмнө
.gitignore 37dc1d74a9 gitignore: per-recipe .lock run ledgers are machine-local state 1 долоо хоног өмнө
README.md f1c95cde2d README: cluster config is the add-on's sm121_config_extra option (no second repo) 1 долоо хоног өмнө

README.md

sm121/recipes — recipe repo (public data repo)

The git recipe repo for the sm121 launcher DGX Spark / GB10 cluster toolkit. The launcher (and the Home Assistant add-on) clone/pull this repo and serve from it:

  • recipes/<name>/recipe.toml — the recipes (source layout, exactly as in the launcher repo)
  • docker-images/<dir>/ — the image build contexts (source of truth; optional when the image is prebuilt in the registry)

Design: specs/tech-architecture/ha-addon-recipe-repos.md in the launcher repo.

No cluster config here (on purpose)

This repo is public — and the cluster config ([[clusters]] with private node IPs, host keys, HA entity names) is not shareable. It is supplied as the HA add-on's sm121_config_extra option (raw TOML — HA option storage is private and on the Pi; no second repo, no PAT).

Supersedes the old publish-bundle registry

This repo previously stored sm121 publish output (flat self-contained bundles at the repo root) as the storage of the public recipe registry (sm121recipes.onlinedb.org). That registry is retired (the add-on recipe-repo flow supersedes it); the bundles were removed on the restructure. sm121 publish (bundles) and sm121 registry (browse/ install) still exist in the launcher but no longer commit to this repo.

Maintenance

Recipes + build contexts move together (the served image must match the recipe revision): update this repo from the launcher repo's tracked state (scripts/seed-recipe-repos.sh in the launcher repo shows the layout).