fix: cap ZeroGPU duration + clearer quota/duration errors

#7
by observantdistressed - opened

Summary

Addresses discussion #6 where users hit opaque errors and cannot see useful logs.

Root cause (from Space logs around the report)

  1. ZeroGPU runs/quota exhausted — conditioner and this Space both returned AppError: ZeroGPU runs limit / quota
  2. GPU duration too largeget_duration estimated 200s+; platform rejected (e.g. 414s / 282s > max)
  3. Occasional corrupt/stale keyframe uploads (UnidentifiedImageError)

Changes

  • Cap @spaces.GPU duration at H3_ZERO_GPU_MAX_DURATION (default 240s)
  • Convert conditioner quota failures to a clear gr.Error
  • Convert duration/quota failures on denoise to gr.Error with tips (Turbo, smaller canvas, shorter duration)

Notes for users

  • Sign in with an HF account that has ZeroGPU quota, or wait for daily reset
  • Prefer Turbo, fast canvases, and shorter durations when quota is tight
observantdistressed changed discussion status to closed

Sign up or log in to comment