Eval Set
Instruction Following Benchmark
Tests whether a model follows explicit formatting and constraint instructions. 12 cases covering format, length, exclusion, and style constraints.
evaluationbenchmarkinginstruction-following
Evaluate the model's ability to follow explicit instructions. For each test case, the model will receive an instruction with constraints. Score based on: - Did it follow the format exactly? (0-3) - Did it respect length constraints? (0-2) - Did it avoid excluded content? (0-2) - Did it satisfy the style requirement? (0-3) **Scoring rubric:** - 9-10: Perfect instruction following - 7-8: Minor violations (e.g. slightly over word limit) - 4-6: Partial following (followed some but not all constraints) - 0-3: Largely ignored the instructions