Implement a constrained generation controller for enforcing output rules. ## Generation Constraints {{constraints}} ## Output Format {{output_format}} ## Enforcement Level {{enforcement_level}} Build the controller: ```python class ConstrainedGenerator: def __init__(self, constraints: List[Constraint]): pass def validate_partial(self, partial_output: str) -> ValidationResult: """Check constraints during generation""" pass def enforce_at_generation(self, logits: np.ndarray, partial: str) -> np.ndarray: """Mask invalid tokens""" pass def post_process(self, output: str) -> str: """Fix constraint violations""" pass ``` Constraint types: - Length limits (min/max) - Format patterns (regex) - Vocabulary restrictions - Structural requirements - Content policies Include: - Soft vs hard constraints - Constraint conflict resolution - Performance optimization
Constrained Generation Controller
U
@
Build a constrained generation controller enforcing length, format, vocabulary, and structural constraints during and after LLM generation.
45 copies0 forks
Details
Category
CodingUse Cases
Constrained generationOutput controlFormat enforcement
Works Best With
claude-sonnet-4-20250514gpt-4o
Created Shared