The training stack now supports a reproducibility seed and an optional use_torch_compile flag (added in #58, plumbed through JobConfig → train_cfg → TrainingConfig). They are settable via the API but not yet surfaced in the wizard UI.
Task: add a numeric Seed input (default 42) and a torch.compile toggle (default off) to the Step 4 hyperparameter form in app/pages/finetune.py, bound to new FinetuneState fields, and include them in the job-creation payload.
Good first issue — the backend and schema already accept both fields.
The training stack now supports a reproducibility
seedand an optionaluse_torch_compileflag (added in #58, plumbed throughJobConfig→train_cfg→TrainingConfig). They are settable via the API but not yet surfaced in the wizard UI.Task: add a numeric Seed input (default 42) and a torch.compile toggle (default off) to the Step 4 hyperparameter form in
app/pages/finetune.py, bound to newFinetuneStatefields, and include them in the job-creation payload.Good first issue — the backend and schema already accept both fields.