openmoji-g1-gpu-cd3250e-0bafd5c-9b9b1699
failed —
Hypothesis
The exact locally verified Gate G pipeline completes one CUDA train step on the owned RTX 4080, restores a canonical checkpoint in durable storage, and preserves locked paths exactly.
Why run it
Validates the real data, selected normalizers, conditioned model, optimizer, checkpoint, and locked-edit path in the target GPU environment before larger work.
Measured behaviour
Table view
| optimizer step | train accuracy |
|---|---|
| 1 | 0.0051 |
Measured result
Verbatim from the run's summary.json.
d11efa006657dafc…0bafd5c3b3de29da…71ee91a4c99b9f53…Written result
Status: failed before model construction or a GPU step.
This run is bounded to the exact one-step local pilot and stages only its six selected raw SVGs plus the pinned palette with the committed source snapshot.
The immutable stage verified, but the pilot resolved its relative input paths against the image working directory rather than the staged source root. It failed on the pinned palette path without writing a checkpoint or result. This run remains preserved; the wrapper working-directory fix requires a new code and run identity.
State transitions
- planned2026-09-20T09:38:14Z
- staged2026-09-20T09:39:28Z
- failed2026-09-20T09:40:06Zstaged_relative_paths_resolved_against_image_workdir
Run record
Verbatim from runs/openmoji-g1-gpu-cd3250e-0bafd5c-9b9b1699/run.yaml, the record committed before launch.
9bcb7145ddd0ec8b…3be6235a1211e440…0bafd5c3b3de29da…9b9b1699677a6f97…e0cdb2a3cc8f00df…