ReJev: LoRA-tuned MiniCPM5-2B reports 80.50% holdout accuracy on bounded decisions
Overview
ReJev, an independent community project, applied LoRA post-training to OpenBMB's MiniCPM5-2B for bounded agent decisions, where a model picks one option from a state, question, and candidate list.
The authors report that accuracy on a sealed 1,892-sample holdout rose from 51.11% to 80.50% (+29.39 points), with 0% invalid outputs and about $5.31 in cumulative Modal billing including earlier experiments.
The authors describe this as an early, task-specific result and explicitly not evidence of parity with Jev.
Written by AI from the articles below · updated Oct 8, 8:55 PM ET
Check the sources:
Article timeline
Follow the coverage from different perspectives. Times are ET.
- OpenBMBReJev fine-tunes MiniCPM5-2B to lift decision accuracy to 80.50%
AIReJev, an independent community project, applied LoRA post-training to OpenBMB's MiniCPM5-2B for bounded agent decisions: state, question, and candidate options yield one choice. On its sealed 1,892-sample holdout, accuracy rose from 51.11% to 80.50% (+29.39 percentage points) with 0% invalid outputs, at about $5.31 in cumulative Modal billing including earlier experimental overhead. The authors describe this as an early, task-specific result, not parity with Jev.
Heat trend
Not enough continuous observations to show a trend yet.