Repository navigation

ARM/CPU inference tier: vLLM on Graviton (PRD-67) - #92

Merged

jicowan merged 10 commits into

mainfrom

feature/cpu-arm-inference-tier

Sep 2, 2026

Merged

ARM/CPU inference tier: vLLM on Graviton (PRD-67)#92
jicowan merged 10 commits into
mainfrom
feature/cpu-arm-inference-tier

Commits on Aug 11, 2026
- feat(runtime): add VLLMcpu ARM/Graviton runtime (PRD-67 §1)
  
  jicowan
  and
  claude
  committed
  6556889
- feat(db): seed Graviton CPU instances + expand accelerator/framework checks (PRD-67 §2)
  
  jicowan
  and
  claude
  committed
  6771abe
- feat(recommend): CPU/ARM host-memory recommender + fix ÷0 dispatch (PRD-67 §4)
  
  jicowan
  and
  claude
  committed
  69125db
- feat(manifest): render CPU/ARM deployments in both templates (PRD-67 §3, §8)
  
  jicowan
  and
  claude
  committed
  4078f19
- feat(karpenter): ARM/Graviton CPU node class + pool (PRD-67 §5)
  
  jicowan
  and
  claude
  committed
  543018a
- feat(cpu): metrics guard + frontend CPU tier (PRD-67 §7)
  
  jicowan
  and
  claude
  committed
  5a15804
- feat(config): settable vllm_cpu_version + wire CPU image version source (PRD-67 §10)
  
  jicowan
  and
  claude
  committed
  0d68a49
- fix(cpu): live-validation fixes — image tag, streamer fallback, loadgen model name (PRD-67)
  
  jicowan
  and
  claude
  committed
  dc8d578
- feat(cpu): server guards + UI gating for CPU-invalid knobs (PRD-67 §11)
  
  jicowan
  and
  claude
  committed
  b5e6354
- test(report): CPU-vs-GPU compare graceful degradation (PRD-67 §12)
  
  jicowan
  and
  claude
  committed
  a5f68c5