2026-10-02
Guided sharpening job failed during tape copy
A guided sharpening task failed while copying results to tape, using one CPU core on a transfer node for four minutes.
What ran
- guided sharpening, saving results ran on a transfer node with 1 CPU core.
- The job started at 2026-10-02T23:55:10Z and ended at 2026-10-02T23:59:20Z, running for 4 min.
- It was allowed up to 8 h but ended with an error (code 1).
What finished
No jobs finished on 2026-10-02.
Numbers
- Jobs that ran: 1
- Jobs finished: 0
- Jobs failed: 1
- Node-hours of ended jobs: 0.0
- Core-hours of ended jobs: 0.07
- red 4x target gain: 0.94 dB over bicubic
- red 4x previous model gain: 0.634 dB over bicubic
- No training runs reported activity that day.
Failures and fixes
- guided sharpening, saving results failed with exit code 1:0. The records do not say why.
- The records list no fixes.
What was learned
- The job "guided sharpening, saving results" ended as failed with error code 1.
- The run sr-v4-fullres-b02 (guided) started.
- In its first epoch, sr-v4-fullres-b02 achieved +6.67 dB over bicubic.
What was decided
- Guided sharpening: Wiring inventory: the learned colour and sharpening readers, their switch rows, and the CPU cost of a sharpened plane.
- Guided sharpening: Full-resolution runs for: Apptainer image per registered commit, array of six runs, timing job, data bundle builder, and the submission guide.
- Guided sharpening: Himawari run, Meteosat run: amendment, the runs execute on with the identical recipe and container; b02 guided resumes from epoch 1.
What is next
- guided sharpening: score the six returned runs against Sentinel-2 on terrain-placed scenes once the independent-truth collection closes, about 10 October
- red 4x sharpening: runs 1 to 5 start when the reserved nodes free up; then choose the best network and loss; runs 6 and 7 follow
- true color v3: a new plan for a day and low-sun hybrid before any further training; the standard recipe stays on the site
- satellite radar global pilot: unpack the data, a 30-minute timing run, then four 72-hour training runs and scoring on the cluster; the Europe test waits for 90 days of radar, about 17 December
- sharpening benchmark: fetch and build the first 20 tiles, read the time per tile, then release the other 880
- coincidence framework: build the one-year census package, about 4 node-hours on CPU nodes
- waiting in the queue: red 4x sharpening, main run (allowed 36 h)
Narrative written by the house model from the day's records and checked by code; decisions and next steps copied from the records. Back to the study page.