kevago/calvino-laya-ft

Calvino Laya fine-tune, run 1

A fine-tune of convaiinnovations/laya (multilingual, commit 7b928d828b7b0e022f929d9bd2e44165aa270148) for Project Calvino's stuck-payments workflow, trained on synthetic, team-generated Spanish customer messages for two questions only: needs_human and workflow_area. Apache License 2.0.

Not for production decisions. The only numbers recorded for this run are training-set fit ("train fit, not evaluation"); whether it is better than the base checkpoint is decided by a separate held-out comparison. See the Calvino repository, TSD-020.