Skip to content

dashboard.md

Latest commit

 

History

History
67 lines (56 loc) · 2.96 KB

dashboard.md

File metadata and controls

67 lines (56 loc) · 2.96 KB
Raw

Feature: Dashboard

Purpose

Give an at-a-glance MLOps overview of fine-tuning runs and the artifacts they version on Backblaze B2, including the cumulative artifact-storage curve.

Used By

  • UI: / page (dashboard home)
  • API: GET /runs/stats, GET /runs

Core Functions

  • apps/web/src/components/dashboard/stats-cards.tsx — 4 stat cards (runs, completed, artifacts on B2, storage on B2), plus the loading notice
  • apps/web/src/components/dashboard/artifact-growth-chart.tsx — area chart of cumulative artifact bytes over time (the accumulating-artifact curve)
  • apps/web/src/components/dashboard/recent-runs-table.tsx — the 8 most recent runs with status and final loss
  • apps/web/src/lib/queries.ts — useRunStats(), useRuns() (both poll while a run is active)
  • services/api/app/service/runs.py — get_run_stats() aggregation
  • services/api/app/service/run_view.py — build_stats() (the cumulative curve)

Canonical Files

  • Pattern exemplar: apps/web/src/components/dashboard/artifact-growth-chart.tsx

Inputs

  • None (reads scoped listings of the app's own runs/ prefix)

Outputs

  • GET /runs/stats → RunStats (totals, status breakdown, cumulative[])
  • GET /runs → RunSummary[] for the recent-runs table (newest first)

Flow

  • Page loads → stats cards read GET /runs/stats (aggregated from each run's manifest plus a scoped runs/ listing).
  • The growth chart plots RunStats.cumulative — one point per run (ordered by completion time) of cumulative artifact bytes, which only grows as runs accumulate on B2.
  • The recent-runs table reads GET /runs and links each row to its detail.
  • All aggregations flow through the runtime -> service -> repo layering and are exposed via TanStack Query hooks (no bare useEffect + fetch). Both queries poll on an interval so a training run's numbers update live.

Edge Cases

  • No runs yet → empty states on every card/chart/table.
  • API unreachable → inline ErrorState with Retry (never fabricated zeros).
  • A run in progress → stats reflect only the checkpoints already written to B2.

UX States

  • Empty: "No runs yet" / "No artifacts yet".
  • Loading: skeletons + a loading notice for the (rare) cold scan.
  • Error: inline error state with Retry.
  • Loaded: populated cards, growth curve, recent-runs table.

Verification

  • Test files: apps/web/src/lib/queries.test.ts, apps/web/src/lib/api-contract.test.ts
  • Required cases: stats render, empty state, error state
  • Focused verify command: pnpm test:web
  • Default pre-PR verify command: pnpm verify
  • Full local verify command: pnpm verify:full when the E2E/live prerequisites in Dev Workflows are available
  • Pass criteria: cards + curve + recent-runs render from live /runs/stats data

Related Docs