Issues

• Scrape pico/dsfront /metrics instead of treating the backend as gauge-less

  MiaAI-Lab/sparkDash#401 · Filter by author robertung88 opened on Nov 13, 2026

• v2.90.0 does not build on case-insensitive filesystems (macOS): ./DeleteAdvice resolves to deleteAdvice.ts

  MiaAI-Lab/sparkDash#389 · Filter by author sgtworkman opened on Nov 12, 2026

• Proposal: optional read-only training and evaluation progress module

  MiaAI-Lab/sparkDash#387 · Filter by author Maxrio9 opened on Nov 11, 2026
  1

• Fleet energy goes blank when one unit is offline

  enhancement
  MiaAI-Lab/sparkDash#359 · Filter by author markyu0401 opened on Nov 2, 2026
  4

• LLM probe: add absolute KV token counts (tile sub-line + history corner), populate them for the tensorfold backend

  MiaAI-Lab/sparkDash#348 · Filter by author de-flandres opened on Nov 1, 2026
  2

• Shared tiered range picker with custom spans, adopted everywhere a time window is chosen

  MiaAI-Lab/sparkDash#347 · Filter by author de-flandres opened on Nov 1, 2026
  1

• Compare view: render the same history panels for the whole fleet on one page/time axis

  MiaAI-Lab/sparkDash#345 · Filter by author de-flandres opened on Nov 1, 2026
  1

• Prefill bench: over-time rate view for a running long-context test + participant line for multi-rank engines

  MiaAI-Lab/sparkDash#344 · Filter by author de-flandres opened on Nov 1, 2026
  1

• KV cache panels: expose the per-pool breakdown in the tooltip (vector read of the worst-lane ratio)

  MiaAI-Lab/sparkDash#342 · Filter by author de-flandres opened on Nov 1, 2026
  1

• Fleet Energy: 'Efficiency' panel charts consumption-per-token (lower is better) under an efficiency name — rename and/or invert to tokens-per-Wh

  MiaAI-Lab/sparkDash#341 · Filter by author de-flandres opened on Nov 1, 2026
  1

• RoCE/RDMA utilization over time: persist per-device RX/TX + loss trends (data already collected)

  MiaAI-Lab/sparkDash#335 · Filter by author de-flandres opened on Nov 1, 2026
  2

• Storage usage over time: historize per-mount usage (worst-mount % / largest-mount used) and add a history chart

  MiaAI-Lab/sparkDash#334 · Filter by author de-flandres opened on Nov 1, 2026
  2

• Record + chart LLM series the probe already computes: avg decode tok/s per stream, cached/uncached prefill split, MTP acceptance

  MiaAI-Lab/sparkDash#333 · Filter by author de-flandres opened on Nov 1, 2026
  1

• Feature: running/queued lanes over time on the LLM history chart (see inflow pressure vs prefill/decode/power)

  MiaAI-Lab/sparkDash#331 · Filter by author de-flandres opened on Nov 1, 2026
  1

• [Feature Request] Serve under a URL sub-path behind a reverse proxy (SPARKDASH_BASE_PATH)

  MiaAI-Lab/sparkDash#192 · Filter by author naeemakhtar110 opened on Oct 11, 2026
  11