GRPO (Group Relative Policy Optimization)

{
  "doc": "natural.md",
  "context": "RL algorithm used to train the activation verbalizer on open models; samples group of candidate descriptions and applies policy optimization.",
  "category": "ai",
  "norm_label": "GRPO (Group Relative Policy Optimization)",
  "graphify_id": "grpo",
  "source_file": "natural.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/natural/graph.json",
  "extracted_type": "method",
  "source_location": "§Method: NLA training",
  "graphify_file_type": "method"
}