LoRA checkpoints of models GRPO-trained on a true or misspecified reward (coding, judge, length), with their training datasets.
-
hewittlab/monitor_cot_for_hack-dataset-coding
Viewer • Updated • 1.98k • 36 -
hewittlab/monitor_cot_for_hack-dataset-judge
Viewer • Updated • 1.11k • 49 -
hewittlab/monitor_cot_for_hack-dataset-length
Viewer • Updated • 12k • 59 -
hewittlab/monitor_cot_for_hack-baseline-coding-explicit_intent-nemotron-3-nano-30b-a3b
Updated