lsrif-logic-structured-rl

Improves instruction-following by using differentiated reward mechanisms that recognize logical dependencies (sequential, conditional, parallel) in instructions, enabling better model reasoning about task structure.

Independent GitHub snapshot · Automated classification · Original documentation governs