Can we get c2606aa merged in?

Aperta
  • Qualcosa non va?
#267 2 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
25/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
rust
Ambito
backend, performance

Direzione di ricerca

Start by reviewing commit c2606aae and its five-commit branch in helios, including the docs and TLA+ spec updates. Compare its io_uring reader lanes with the approach in closed PR #216 and determine whether that earlier rejection applies here. The change is behind --embedding-table-prefetch and --embedding-table-workers; validation should include the reported prefill metrics and measurement on real DeepSeek hardware, which the issue says is still outstanding.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

I'd propose topic/engramcache (head c2606aae, five commits) to be merged into main. On disk-backed Engram tables, prefill stalls because a graph publishes its row ids only when it starts, so each miss is read while the GPU waits, and the prefetch meant to hide that loaded nothing: one 4x RTX PRO 6000 run logged 65 submissions, 0 rows loaded and 5,568 dropped. The branch holds a job until its ids are readable, pins read-ahead rows so they can't be evicted before use, and gives each reader worker its own io_uring lane. A host emulation at DeepSeek V4.1's table shape on a GB10 NVMe measured 10.8k vs 15.1k tokens/s and 60% vs 84% GPU busy on a cold 256k prompt with a 16 GiB cache. Real DeepSeek hardware is still unmeasured. The whole series sits in helios behind --embedding-table-prefetch and --embedding-table-workers and carries its docs and TLA+ spec updates. PR #216 tried the same io_uring lanes and was closed unmerged, so it would help to know whether it was rejected for something this branch does too.

https://github.com/randomvariable/spark-local-inference/commit/c2606aae

Lingua principale
Rust
Stelle
800
Fork
106
Merge medio
1g 3h
PR unite (30g)
100

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di NVIDIA/hyperflare-inference

Tutte le issue di NVIDIA/hyperflare-inference

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.